Skip to main content
Internet Archive's 25th Anniversary Logo

Full text of "New Welsh consonant shift, description and implications"

See other formats












I Dduwies Gobaith 


I would like to thank the members of my supervisory committee, 
Professors Bohdan Saciuk, Ghauncey Chu, Donald Dew, Eileen Sullivan, 
and William Sullivan, for the help and enthusiasm they have shown me 
in the preparation of this dissertation. I would also like to thank 
Dr. T. Arwyn Watkins of the University College of Wales at Aberystwyth 
for the comments he has given me and the interest he has shown in this 
work. For their patience and cooperation, I owe a debt of gratitude 
to my informants, Dr. Bedwyr Jones, Mr. Merfyn Morgan, and Mr. Idris 
Roberts of Bangor; Mr. Hugh Jones of Bettws Gannon? and Mrs. Mona 
Pringle of Gainesville. Unfortunately, space does not permit me to 
thank by name the many professors both here and in Wales who have 
given me invaluable guidance in my research in linguistics and in the 
Welsh language. Nor could I list in this short space the names of my 
fellow students who have given me advice and ideas which have added 
greatly to the depth of my work. I would like to thank my wife, Donna, 
and my relatives and friends whose confidence in me has made my work 
considerably easier. This dissertation could not have been written 
without the help of my parents, who gave me the foundations of my 
education in sending me to The Citadel. 



The purpose of this work is twofold— to describe certain consonant 
phenomena in Welsh which may be indicative of a consonant shift in 
progress, and to introduce a nonsegmental phonology, termed hierarchical 
phonology. In describing the consonant phenomena, I use the trans- 
formational generative approach to synchronic and historical descrip- 
tion, as this approach is perhaps the most widely used model in lin- 
guistics today. 

The proposed hierarchical phonology is presently in the initial 
stages of development. In order to maintain the flexibility necessary 
to allow the model to develop freely, I use the basic functional/struc- 
tural theoretical approach characteristic of the Prague school, especially 
as it is found in the works of Trubetzkoy and Jakobson. To this basic 
structure, I add the findings of experimental phoneticians, such as 
Mermelstein and <3hman, who have recently developed dynamic phonetic 
models for the study of speech production. 

In order to describe the consonant phenomena in Welsh, I first 
must describe the phonology of the consonant subsystem, especially as 
regards the 'mutation system', a system of initial consonant gradation. 
Chapter 1 describes the phonological aspects of the system, while 
Chapter 2 treats the grammatical motivation associated with the muta- 
tions. Chapter 3 then introduces an area in which the otherwise 
regular system appears to become highly irregular. The irregularity 
of the system in the 'deviation' is attributed in Chapter 4 to a case 


of historical change in which voiceless aspirated stops appear to be 
changing to voiced unaspirated stops. Further examination of these and 
related phenomena in Chapter 5 gives strong indications that the changes 
exhibited in the mutation system are indicative of a consonant shift and 
that the shift is presently in progress in the Welsh language. In 
Chapter 6, however, we find that the conclusion that a consonant shift 
is in progress in Welsh is based upon the acceptability of the generative 
description and that the generative description is in many ways faulty. 

The hierarchical model is then proposed in an attempt to provide 
a more reliable description of Welsh and to incorporate recent evidence 
from phonetics into phonology. The notion that aspiration, a prosodic 
opposition, forms the basis of the Welsh consonant subsystem is intro- 
duced in Chapter 7 and supported by data from the spoken and written 
language. Using this prosodic gradual opposition as a base raises 
doubts as to the validity of a segmental phonology. Coupling these 
doubts with the phonetic dynamic models in which speech sound is described 
without recourse to segmentation, I then construct a basic nonsegmental 
model, or hierarchical phonology, in Chapter 8. This model is applied 
to the phenomena in Welsh in Chapter 9. Thus, Chapter 9 is the conclusion 
of the work, combining the hierarchical phonology with the consonant 

At the center of this hierarchical phonology is the notion that the 
phonology must reflect the evidence of phonetics. In keeping with this 
notion, Chapter 10 examines the evidence from acoustic and physiological 
phonetics supporting the prosodic gradual opposition of aspiration, upon 
which the consonant subsystem of Welsh is based, at least as it is 
presented in Chapter 9. The importance of establishing the phonological 


structure of phonetic evidence cannot be overemphasized. By a carefully 
controlled procedure of abstracting only justifiable phonetic charac- 
teristics as oppositions, we canmaintain a structure which is not only 
consistent but also reliable, at least as far as our current knowledge 
of phonetics will allow. 

In order to facilitate understanding in the midst of frequent 
cross-references, I use a reference system for rules, figures, tables, 
etc., based upon the section numbers. For example, a rule referred to 
as rule 1.3 is the rule found in section 1.3. Where more than one such 
rule is found in a section, I use letters, such as rule 1.1. a, rule 
l.l.b, etc. This system will hopefully prove more helpful than one 
using simply numbers in sequence. 


Acknowledgments l y 

Preface v 

Abstract x 

Chapter 1, The Phonology of Mutation 1 

1.0 The Mutation System 1 

1.1 Soft Mutation 7 

1.2 Nasal Mutation 13 

1.3 Spirant Mutation 16 

1.4 Aspirate Mutation 16 

1.5 Functionality and Redundancy 17 

1.6 The Phonology of Mutation 20 
Notes to Chapter 1 22 

Chapter 2, The Grammar of Mutation 24 

2.0 Grammatical Motivation for Mutation 24 

2.1 Environment I 28 

2.2 Environment II 31 

2.3 Environment III ™ 

2.4 Environment IV ^9 

2.5 Environment V 54 

2.6 Overlapping Environments 55 

2.7 Constituent Environments 57 

2.8 The Totality of the Environments 60 
Notes to Chapter 2 63 

Chapter 3, The Deviation— Synchronic Analysis 65 

3.0 Exceptions and Counter-Examples 65 

3.1 The Problem of gan 69 

3.2 The Minor Rule Based upon I gan I 71 

3.3 The Minor Rule Based upon (kan| 76 

3.4 Evidence for the Minor Rule 80 

3.5 The Synchronic Description 83 
Notes to Chapter 3 8 5 

Chapter 4, The Deviation—Historical Analysis 86 

4.0 The Generalization of the Minor Rule 86 

4.1 The Historical Development of the Minor Rule 89 

4.2 The Generalization to the Natural Class 95 
Notes to Chapter 4 98 



Chapter 5, A New Welsh Consonant Shift 99 

5.0 Sound Change and Consonant Shifts 99 

5.1 The Position of Neutralization 104 

5.2 Medial Positions and Clusters 111 

5.3 Initial Position l* 8 

5.4 Quod Erat Demonstrandum 124 
Notes to Chapter 5 12 ? 

Chapter 6, Problems of Description 128 

6.0 Problems in the Generative Description 128 

6.1 The Inconsistencies of Soft Mutation 134 

6.2 Markedness and the Minor Rule 140 

6.3 Directions for Phonological Inquiry 148 
Notes to Chapter 6 151 

Chapter 7, Aspiration in Welsh 152 

7.0 The Basic Opposition of Welsh Consonants 152 

7.1 The Prosodic Nature of Aspiration 157 

7.2 Aspiration in the Phonological System 165 

7.3 The Implications of a Prosodic Base 182 
Notes to Chapter 7 l8 5 

Chapter 8, An Hierarchical Phonology 186 

8.0 The Phonetic Basis of Oppositions 186 

8.1 The Organization of Oppositions 192 

8.2 An Hierarchical Structure in Phonology 204 

8.3 Toward an Hierarchical Phonology 211 
Notes to Chapter 8 217 

Chapter 9, An Hierarchical Description 218 

9.0 Theoretical Implications in the Application 

of the Model 218 

9.1 The Welsh Consonant Subsystem 222 

9.2 Mutation, Lenition, and Provection 233 

9.3 The Deviation 245 
9.^ The New Welsh Consonant Shift 250 
Notes to Chapter 9 2 ° 3 

Chapter 10, The Phonetic Basis 265 

10.0 Phonetic Justification 265 

10.1 Acoustic Basis of Aspiration 268 

10.2 Physiological Basis of Aspiration 280 

10.3 Behavioral Justification of Aspiration 284 
Notes to Chapter 10 290 



Biographical Sketch 302 

Abstract of Dissertation Presented to the Graduate Council 
of the University of Florida in Partial Fulfillment of the Requirements 
for the Degree of Doctor of Philosophy 



Toby D. Griff en 

June, 1975 

Chairman! Bohdan Saciuk 
Major Department i Linguistics 

In a generative description of the Welsh consonant subsystem, 

including the 'mutation system' (a system of initial consonant gradation), 

some irregularities in an otherwise regular system indicate that there may 

be a consonant shift in progress in which voiceless aspirated stops are 

changing to their voiced unaspirated cognates. In examining the evidence, 

however, several problems are found in the description, problems stemming 

from the nature of the generative description. Developments in phonetics, 

especially the recent physiological and acoustic dynamic models which 

describe speech without segmentation, are then combined with a reanalysis 

of Welsh based upon the prosodic gradual opposition of aspiration, in 

order to create a nonsegmental approach to phonological description, 

termed hierarchical phonology. This hierarchical phonology is used to 

provide a more regular description of the Welsh consonant subsystem, a 

description more closely based upon findings of experimental phonetics. 



1.0 The Mutation System . It is a property characteristic of Celtic 
languages that the initial consonant of a word will vary depending upon 
the grammatical, that is, morphological and syntactic, context affecting 
the word (compare Lewis and Pederson 1937). This system of initial con- 
sonant gradation is termed the 'mutation system*. 

Through its relationship with grammatical contexts, the mutation 
systems of the Celtic languages function in much the same way as the 
inflectional systems of languages such as Latin, For example, In 
Latin an adjective modifying a first declension feminine noun in the 
nominative case will have to agree in number, case, and gender with 
that noun. In Modem Welsh, on the other hand, the adjective modifying 
a feminine noun will undergo a particular mutation — a change in the 
initial consonant. Furthermore, in Latin a noun which is the object of 
the verb will be marked by a particular inflectional ending, usually the 
accusative. In Welsh, however, the object of an inflected verb will 
undergo a particular mutation of the initial consonant. 

Throughout the Celtic languages, the changes of initial consonants 
reflect a certain systematicity. One type of mutation may be thought 
of as exhibiting a process of lenition, another of nasalization, and 
another of frication. As I demonstrate below, these changes are indeed 
regular applications of processes representing certain phonetic charac- 
teristics (or, at least, they can be described in this manner). These 

mutations are, then, as Hamp 1951 points out, morphophonological. 

In Modern Welsh, the mutations are dependent upon the grammatical 
context. Although the system is now morphophonological, it developed 
from phonological alternations that existed during the transition from 
Brythonic to Old Welsh (up to about the end of the eighth century — 
Morris Jones 1913*6). For example, when Brythonic */m/ occurred between 
vowels, it was realized in Old Welsh as /p/, a voiced bilabial nasal 
fricative, in Modern Welsh as /v/. At the same time that these phono- 
logical alternations were taking place, the final syllables of Brythonic 
words were being lost, Including the feminine ending */a/. Thus, the 
*/m/ in Brythonic *oinos markos */o±nos markos/ 'one horse' is realized 
as /m/ in Modern Welsh un march /in marx/, as there is no vowel before 
the */m/ to cause the phonological change in the Brythonic word} while, 
on the other hand, the */»/ in Brythonic *oing mamma */o ina « Jaammai/ 
•one mother* is realized as /v/ in Modern Welsh un fam /in vam/, as 
there is a vowel before the */m/ which caused Brythonic */»/ *° be 
realized as Old Welsh /!*/, Modern Welsh /v/ (see Morris Jones 1913 «l6l). 

Although the phonological context was sufficient to bring about 
alternations in the Brythonic-to-Old Welsh period, it is not sufficient 
(nor even necessary) to cause alternations now. The examples above 
should show that there is no phonological motivation for the change in 
Modern Welsh, 

The usual way of depicting the mutation system in Modern Welsh in 
the standard orthography (see Bwrdd Gwybodau Celtaidd 19^-2), is presented 
in Table 1.0, a. The phonological segments corresponding to the ortho- 
graphy in Table 1.0. a can be found in Table 1.0. b. 

The left-hand column in the tables contains the segments known as 

•radicals'. The radical segment is the one listed in the dictionary. 
For example, the word 'mother* in the example above would be entered 
as mam, regardless of the fact that it is realized in that particular 
phrase as fam. 

Whenever none of the mutation rules apply, the radical is the 
segment realized. Moreover, where a blank space occurs in the tables, 
the radical is realized. For example, in the grammatical context of 
spirant mutation, mam /mam/ would simply be realized as mam /mam/, 
there being no spirant mutation form of /m/. 

Within these initial chapters (1 through 5), in which I describe 
the new Welsh consonant shift and present arguments for it, I utilize 
basically the transformational -generative framework, as found in Chomsky 
and Halle 1968. The notation which I use, however, is that of Schane 
1973, as this notation is compatible with the notations of most other 
schools of linguistics and it allows us to differentiate between a 
phonological segment (in slashes--/ /) and a phonetic segment (in 
brackets~[ "]) The underlying segment is enclosed in vertical bars 

(I I). 2 

In the generative framework, the mutation system of Modern Welsh 
is expressed as a series of rules which derive a phonetic segment from 
an underlying segment. This item-and-process framework (see Hockett 
1954) assumes, then, that there is an item at an abstract 'deep' level 
which is transformed through orderly processes into an item at the 
•surface' level. Thus, for every line in the mutation tables, there 
is an underlying segment (not to be confused with the morphophoneme of 
Hamp 1951, which is an abstraction of surface oppositions), and this 
underlying segment undergoes various process rules in order to generate 
the surface phonetic segments. 

Table 1.0, a 
The Mutation System — Othography 







ant mutation 




























Table 1.0. b 
The Mutation System— Phonological Transcription 






spirant mutation 







e* c 

















l a 




a /l/ is a voiceless lateral fricative. 

V*/ is a voiceless trill. 

CThese are voiceless aspirated nasals. 

In the case of the mutation system, I consider the underlying 
segment to correspond to the radical. There are several reasons for 
positing the radical as the underlying segment. First of all, the 
radical is the dictionary form, considered by the native speaker to 
"be the basic form. It is the Welshman's intuition (and the traditional 
way of teaching the system) that the mutation forms are derived (via 
•mutation') from the basic radical. In fact, according to Professor 
Gelnwen Thomas of the University of Wales at Cardiff (personal communica- 
tion), when a Welshman is shown the word phen /fen/ (spirant mutation 
form of p_en /pen/ 'head')-* without the grammatical context, he is 
generally at a loss as to how to translate or define it, until it is 
put in its radical form. 

In addition to the intuition of the native speaker, there are 
also distributional considerations for treating the radical as the 
underlying form. As stated above, where there is no nasal or spirant 
mutation form, it is the radical which is realized. Moreover, the 
radical is realized in the absence of a rule within the system itself. 
Thus, it would appear that the radical is the underlying form, as it 
is the neutral segment (unmarked in the Prague sense of the term — see 

Trubetzkoy 1969). 

As we turn our attention to considerations more closely associated 
with generative methodology, we can see that the various mutations 
can be derived from the radical with fewer feature changes than they 
can from any other mutation form. For example, the nasal and spirant 
mutations share more features (per line in the tables) with the radical 
than they do with the corresponding soft mutation segment. Thus, 
greater simplicity in phonological rules can be realized if we posit 

the soft mutation forms as underlying. 

Perhaps the best motivation for positing the radical as the under- 
lying segment is the fact that all mutations can be predicted from the 
radical, but the radical cannot be predicted from all mutation forms. 
For example, there are two occurrences of /v/ in the soft mutation, 
one of which corresponds to the radical /b/ and one to the radical /m/ 
(compare Fowkes 19^9«208; Griff en In press a). Given underlying |v|, 
then, we would be at a loss as to whether to derive /b/ or /m/ in the 
radical environment. Given underlying |b| and |m|, however, we would 
have no problem deriving /v/ in the soft mutation environment. This 
is an important consideration, for the generative model has formal, 
explicit rules operating from deep to surface structure, but not from 
surface to deep, as shown in Chomsky and Halle 1968i29^ (see also 
Griff en In press a). As such, then, cases of diversification which 
cannot be predicted in this formalism are to be avoided at all costs, 
and only the radical can assure us of no diversification. 

Temporarily leaving the generative framework, if we were to view 
the phonology as a series of abstractions from the acoustic data, then 
we might consider this underlying form to be the most basic segment. 
In the terminology of Trubetzkoy (l969tGhapter 4j compare also Venne- 
mann 1972), it is the least marked segment with respect to the distinc- 
tive oppositions (such as voice, nasality, and continuance). In such 
a framework, the underlying segment (the morphophoneme) would become 
marked for various distinctive features in the process of the phonological 
rules. Of course, the one feature (opposition) that is constant through- 
out this process of marking the basic segment is the phonological 
position of articulation. 

This notion of abstraction is not a part of generative phonology 
as found in Chomsky and Halle 1968. Nevertheless, it is a useful way 
of viewing the mutation system. Of course, in a rigorous generative 
description, we can only use the analysis-by-synthesis methodology 
developed in Halle and Stevens 1964 and Chomsky and Halle 196 8. 

As this is a generative description, the rules and segments used 
herein must be specified by distinctive features. I therefore list the 
features and their specifications for the consonants of Cymraeg Safonal 
(that is, the Standard Welsh— see, for example, Watkins I96I) in Table 
l.O.c. The features used are those found in Chomsky and Halle (19681 
Chapter 7). Instead of heightened subglottal pressure, however, I 
use aspiration. The justification for using aspiration as a distinctive 
feature in addition to voice in Welsh is provided in Chapter 7. The 
feature trill is necessary as a means of distinguishing /9/ from /*/. 
I do not include the affricates, and I do not deal with them in this 
study. These are, however, treated in Griffen 197^b (see also R.O. 
Jones 1971X 

1.1 Soft Mutation . As we can see in Table 1.0. b, the soft mutation, 
also known as lenition, cannot be characterized by any one single 
feature change using the binary features of Chomsky and Halle I968. 
In fact, the soft mutation cannot be described within the generative 
framework as a single phonological rule. 

Nevertheless, there is a degree of phonetic unity within this 
process, because each rule in the soft mutation serves to create a 
segment which is more lenis (or soft— see Malmberg 1963»52) than the 
underlying segment. In strictly binary features, this lenition process 



I i 


i i 


i i 


+ + 


+ + 



+ i 



+ i 



+ I 



+ 1 



+ i 



+ i 



+ 1 



+ i 


+ i 





+ 1 




+ i 

O <H 

• o 



• m 


+ 1 

•rH +> 

4) 3 




+ 1 

«5 to 

H S 


+ 1 





+ 1 





+ 1 



+ 1 



+ I 



+ 1 



■f 1 










1 + 

















































+ 1 









































1 1 








1 t 




1 + 





1 + 




1 + 



























o 3 

O S3 







must be described through other binary features (compare Cherry, Halle, 
and Jakobson 1953 j Vennemann and Ladefoged 1973). I return to this 
issue in section 6.1, 

The soft mutation can be divided into five generative phonological 
rules. These are posited as follows » 

a. fpl - /b/ r. T\ 

f-vcdl _ Rvcd| / +obs 
It I - /d/ L-faspJ L-aspJ ' L- cnt J 

Ikl- /g/ 

In other words, all voiceless aspirated stops become voiced and unas- 
pirated. Of course, this rule assumes the proper grammatical context, 
which I posit in detail in the next chapter.5 

The advantage of the feature specification notation in phonological 
rules is that it allows us to capture a generalization. In this case, 
voicing and deaspiration take place in this class of segments under 
soft mutation regardless of the position of articulation. Moreover, 
this rule demonstrates a change in an entire natural class. 

The notion of a natural class is central to this description. 
According to Harms (I968i26), there are two considerations involved 
in the notion of a natural class i 'First, it is a class of segments 
that can be specified with fewer features than any individual member of 
the class,.,. Second, the features shared by the class members should 
be limited to those which have a certain degree of phonetic plausibility. 
Recall that in rule 1.1. a we do not specify any features of position of 
articulation (such as back, anterior, and coronal), because the rule 
affects all stops regardless of position. Thus, the first consideration 
of a natural class is satisfied. As for the second condition, there 


should be little argument as to the plausibility of the class of voice- 
less aspirated stops within the phonetic system. This class has been 
one of the traditional classes in linguistic description—the aspiratae 
(as opposed to the tenues and mediae). 

Within the generative framework, then, the voiceless aspirated 
stops (aspiratae) function as a single class. In the remaining soft 
mutation rules, however, we cannot form such a neat generalization 
within this framework. 

The voiced unaspirated stops undergo the following changes: 

b. Ibl - /v/ C n 

C+obsD - C+cnfl / hvcd 
|d| - /6/ l-bckj 


c igi - $ K cd ■* i 


There are two aspects of rules 1.1. b-c which are somewhat dis- 
concerting. The first (and more minor) is that there appears to be 
a connection in this rule between continuance and nonbackness. That 
is to say that we might want to find a relationship between the notion 
of continuance and that of nonbackness. What relationship there may 
be, however, is purely a negative one — nonbackness is introduced merely 
as a means to block the rule from applying to | g | . 

The other disconcerting aspect of these rules is that fact that 
fb|, |d I, and lg| are members of a natural class but they do not 
undergo a common rule. This problem has been approached by Zwicky 197^, 
in which an intermediate /y/ is posited. This allows for the natural 
class to undergo a common rule, but it adds another rule to the grammar— 
/y/ must be deleted wherever it occurs. Now historically, there was 


indeed a /y/ as the soft mutation form of \g\ (Morris Jones 1913il6l). 
Resurrecting such a segment merely for the purpose of deleting it, 
however, is a questionable practice. We must bear in mind that in the 
generative item-and-process framework, such an intermediate form is an 
item in the derivation, not simply a place-holder to demonstrate the 
incompatibility of certain features, and as an item its creation must 
be fully justified. 

Returning once more to the initial disconcerting aspect, the use 
of nonbackness to block the application in one member of a natural 
class, we can see a bothersome question in the absence of the historical 
/y/. As the inclusion of the bilabial and the apico-dental voiced 
stops in a single rule requires fewer features and the pair are clearly 
to be classed together, is what we have truly a natural class? Is the 
class of /b/ and /d/ without /g/ an entire natural class? Clearly the 
segments represented in rule 1.1. b do not represent as much of a natural 
class as those represented in rule 1.1. a. 

The next soft mutation rule is as follows » 

d. Ul- hi r _ _ _ 

pens] _> j+cnt| j 

L-vocJ L-nasJ 

As mentioned above, historically there was an intermediate stage 
in which /&/ is used. Although we could gain simpler rules from a 
stage using /&/ between |m| and /v/, such a stage would be as unmotivated 
as the proposal for using /y/ in our synchronic description. Note 
should be taken that through soft mutation, |b| and 1ml neutralize into 
/v/. This fact could prompt us to write a rule whereby voiced labials, 
regardless of nasality, become fricatives. This would, however, complicate 


the rules, as a rule affecting |d| would have to be included separately 
(unlike I mj, Jnj undergoes no mutation). 

Finally, there is the following soft nutation rulei 

e * W" N r 1 n. i / 

L+conJ -* PWocj / 

M- hi bvcdj 


Here again, there is a problem with natural classes. Although the 
output of rule 1.1. e clearly forms a natural class, the class of liquids, 
the input does not. It is difficult, then, to make any insightful 
statement about the natural class involved in this rule. 

Rule 1.1. e makes use of a notational device known as 'braces'. 
The question needs to be asked as to whether this is one rule or two. 
By leaving features unspecified, such as the position of articulation 
features in 1.1. a, we collapse rules into a more general statement. 
The braces, on the other hand, do not represent a generalization, but 
stand for the logical exclusive 'or' relationship. As each application 
of each rule is a singulary entity in a temporal (linear) plane, it 
could be argued that the effect of rule 1.1. e is that of two rules, 
and the method of collapsing the rules is merely a notational convenience 
rather than a statement of generalization. 

According to Chomsky and Halle (I9681333), 'two partially identical 
rules may be coalesced into a single rule by enclosing corresponding 
nonidentical parts in braces.' As the interpretation of Chomsky and 
Halle is clearly that the braces collapse two rules into one, I treat 
rule 1.1. e as a single rule. The objection, however, ought to be con- 
sidered from a logical viewpoint. 

Thus, given the notion that rules 1.1. b (in spite of the fact that 


it implies a relationship between continuance and nonbackness and works 
to exclude a natural class) and l.i.e (in spite of the logical considera- 
tions regarding the braces) each represent a single rule, the soft 
mutation of Modern Welsh can be represented in five rules. I do not 
believe that a credible argument could be made for collapsing the rules 
any further, as any simplicity that could be gained from collapsing the 
rules so as to create fewer rules would in turn be lost in the extreme 
complexity of the rules themselves. 

So far, I have not given any examples of the application of these 
mutation rules. The rules are intricately connected with their gram- 
matical contexts, and I give examples of them in Chapter 2, where I 
address these contexts. 

1.2 Nasal Mutation . The soft mutation is described in five generative 
rules. Even if these could be collapsed further, the fact would remain 
that in the feature system of Chomsky and Halle 1968, which is neces- 
sarily binary, there are five processes involved—voicing, continuance, 
deletion, continuance and denasalization, and vocalization and voicing. 
The nasal mutation, on the other hand, can be described in only one 
rule. This rule can be written as in 1.2.c, below. 

We should notice that in rule 1.2.C, the voiceless aspirated stops 
(aspiratae) and the voiced unaspirated stops (mediae) together form the 
natural class of all stops. This is reflected in the notation by the 
use of a minimum of features. Nonetheless, I split the rule into two 
rules in order to maintain the notion that voiceless aspirated stops and 
voiced unaspirated stops are indeed two different natural classes in 
the operation of the mutation system (compare also section 1.3) » and the 


nasal mutation, according to Professor Robert Owen Jones of the Univer- 
sity College of North Wales at Bangor (personal communication), can be 
thought of as aspirated stops becoming aspirated nasals and unaspirated 
stops becoming unaspirated nasals (see Griff en 197^b»159). 

The two rules and the collapsed single rule are written as follows i 


IpI — 


1*1 - 


Ikl - 



Ibl - 


Idl - 


lei — 


[-nas] •* [-hias] / 

[-nas] — [+nas] / 







[-nas] - [+nas] / 

r vocJ 

As Modern Welsh has no segments that are both nasal and continuant 
(/f/ is found in Old Welsh but changes to /v/, as mentioned above), 
there is no need to specify continuance in these rules (at least by the 
feature criteria of Chomsky and Halle 1968), 

As stated above, there is certainly a natural class of stops 
(though it must be subdivided into two classes in soft mutation), and 
this natural class becomes nasal in the nasal mutation. In effect, then, 
there is only one nasal mutation rule (l.2.c). On the other hand, the 
voiceless aspirated stops and the voiced unaspirated stops act quite 
differently in the other mutations. Now this fact does not make all stops 
less of a natural class in the nasal mutation, but it does necessitate 
our examining the two classes of stops separately in a number of operations 


in which they could be considered a single class. Moreover, it becomes 
necessary to refer to the aspiration of the nasal segments in 1.2. a 
as distinct from the lack of aspiration in 1.2.b, especially as this 
aspiration may be vital in the productive nature of certain mutations. 

One other aspect of the nasal mutation which becomes important in 
this study should be mentioned. The voiceless aspirated nasals are only 
found as a result of mutation— they are nowhere found as underlying 
segments. The aspiration involved, moreover, is not the same degree 
as that found in the underlying voiceless stop. In fact, it is so 
strong that R.O. Jones I969 describes it as a glottal fricative. This 
fact should be kept in mind, as it is particularly crucial to the argu- 
ments presented (see Chapter ?). 

This heightened aspiration is always realized as the pure [h] 
fricative, even if the corresponding stops are (noncontrastively) 
affricated. For example, Mrs. Mona Pringle of Gainesville, Florida, is 
a native speaker of the Colwyn Bay dialect of North Welsh. Although her 
English /t/ is aspirated [t h ],her Welsh /t/ is affricated [t s ]. None- 
theless, this affrication has no effect upon her nasal mutation forms, 
which are realized with the heightened aspiration [h]. 

As pointed out in Griff en 1975 » the aspiration of the voiceless 
stop is sufficient to devoice the following liquid but not the following 
nasal. The heightened aspiration of the nasal mutation form, on the 
other hand, is sufficient to devoice the following liquid or nasal, and 
as shown in section 7.1, it continues into the vowel following the 
liquid or nasal. Thus, for example, in the Bangor dialect penelin 
[pnelin] 'elbow' is realized in nasal mutation as mhenelin [mnhelin] 
(see Fynes-Clinton 1913 »^36), and that of pleth [pjeie] is [mlhe»e] 

(Fynes-Clinton 1913 «^32). 

1.3 Spirant Mutation . The final mutation which affects consonants is 
the spirant mutation. This mutation also has the most restricted range 
of application, affecting only voiceless aspirated stops. The spirant 
mutation rule is posited as follows i 

IpI- M 

|t|- /©/ [+obs] - [-tent] / pvc(f| 

lkl- /x/ 

At this point, I should stress the regularity with which the 
voiceless aspirated stops have been treated in the mutation system. 
Whereas the voiced unaspirated stops do not operate as an entire 
natural class in the soft mutation and are not even affected by the 
spirant mutation, the voiceless aspirated stops always operate as a 
natural class— in rules 1,1, a, 1,2, a (and 1.2,c), and 1,3. As a result 
of this regularity, the class of voiceless aspirated stops is described 
within the generative phonological framework with the maximum economy 
and generality, as each change affects one and only one feature in one 
natural class, and every mutation affects the class, 

1.4 Aspirate Mutation . In Standard Welsh, the aspirate mutation is not 
commonly considered to be a part of the consonant mutation system, as 

it does not affect consonants. Under the conditions of aspirate mutation, 
an initial vowel undergoes preaspiration. In the generative framework, 
the only way to describe such a phenomenon as preaspiration of an initial 
vowel is to posit a rule whereby the glottal fricative /h/ is inserted in 


initial position, as follows: 


■* -ens / # r+voc"| 

^hgh] " L-cnsJ 

Although not in Standard Welsh, as Fynes-Clinton (l913«xviii) 
points out, the aspirate mutation does in fact affect /ra/, /y/, and 
/w/ in the Bangor dialect.? The extension of aspirate mutation to 
consonants is treated further in Chapter 7. 

1.5 Functionality and Redundancy . As stated in section 1.0, the 
mutation system of Modern Welsh corresponds roughly to an inflectional 
system in such languages as Latin. As is true with many inflectional 
systems, such as that of Latin, the elements of the system can reflect 
various degrees of functionality and various degrees of redundancy. 

The functionality of the soft mutation rules may sometimes be of 
a relatively low degree. For instance, a sentence may be preceded by a 
sentence marker fe /ve/ or ml /mi/ (the choice being largely determined 
by dialect). When this happens, the initial segment of the following 
verb undergoes soft mutation. Thus, a sentence such as gwelodd ef dad 
/gwelod ev dad/ 'he saw a father' may optionally be preceded by the 
indicative sentence marker fe /ve/, in which case, through the soft 
mutation rule l.l.c, above, we obtain fe welodd ef dad /ve weloo ef dad/. 
The functionality of the rule in this case can only be considered 
minimal, as the sentence conveys exactly the same meaning with or with- 
out the addition of the fe /ve/ and its subsequent triggering of the 
soft mutation rule. 

On the other hand, the functionality of the soft mutation may be 


of a relatively high degree. For example, a noun which is the direct 
object of an inflected verb undergoes the soft mutation of its initial 
consonant. In the sentence gwelodd ef dad /gweloo ev dad/, introduced 
above, the initial segment of dad /dad/ is derived by way of the soft 
mutation rule 1,1, a from underlying |t| in tad /tad/ 'father'. Thus, 
in this sentence the soft mutation functions to designate the direct 

As far as redundancy is concerned, in the sentence gwelodd ef dad, 
there is a high degree of redundancy in the soft mutation form of the 
initial segment in dad /dad/. The information that dad /dad/ is the 
direct object of the inflected verb is not only conveyed by the soft 
mutation form, but it is also conveyed by the position of the word in 
the sentence. On the other hand, if we were to delete the redundant ef 
/ev/ (redundant because of the third person singular ending on the verb), 
we would be left with the sentence gwelodd dad /gweloo dad/. In this 
latter sentence, the soft mutation of the initial segment in dad /dad/ 
is the only indication that the noun is the direct object of the inflected 
verb ('he saw a father') rather than the subject of the verb ('a father 
saw (something)'). Thus, in this latter sentence, the functionality is 
of a high degree. 

Moreover, the third person singular possessive pronouns for masculine 
and for feminine gender are identical in phonological shape. The only 
way that the speaker has to tell them apart (other things being equal) is 
by the fact that the masculine gender governs the soft mutation in such 
instances and the feminine gender governs the spirant mutation. For 
example, ei dad /i» dad/ means 'his father', and ei thad /ii 9ad/ means 
•her father'. 


Usually, the mutation forms are redundant to some degree. Because 
not all segments undergo mutation rules (for example, /s/ takes no part 
in any mutation), it would necessarily have to be the case that some 
other grammatical means should also be able to convey the information 
conveyed by mutation. For example, the nasal mutation rules are 
motivated by the first person singular possessive pronoun. Thus, fy_ 
nhad /va nhad/ 'my father' is derived by rule 1,2. a (l.2.c) above. The 

— — 

same grammatical information (that is, first person singular possessive) 
conveyed by the nasal mutation of the initial segment is also conveyed 
by the pronoun itself. Now in this case, the pronoun may be deleted, 
reducing the redundancy of the nasal mutation form. However, when the 
possessed word does not begin with a segment that undergoes nasal muta- 
tion, the pronoun cannot be deleted without the loss of meaning (other 
things being equal). For example, in fy_ mam /vs mam/ 'my mother', the 
initial segment of the noun does not undergo nasal mutation. Thus, the 
pronoun cannot be deleted without losing the information of the first 
person singular possessive. 

In the case of the third person singular possessive pronouns, 
where both masculine and feminine have the same phonological shape and 
differ only in their mutations, there are other ways of telling the two 
pronouns apart when the initial segment of the possessed noun does not 
undergo either soft or spirant mutation. For example, the noun chwaer 
/xwatir/ 'sister' does not undergo any mutation. When this word is 
possessed by either the third person singular possessive masculine or 
feminine pronoun, a form of the pronoun follows, as in ei chwaer ef 
/it xwaiir ev/ 'his sister' and ei chwaer hi /ij xwa:ir hij/ 'her sister'. 


1,6 The Phonology of Mutation . There are two sides to the notion of 
the Welsh mutation system — a phonological side and a grammatical side. 
In this chapter, we are only concerned with the phonological considera- 
tions inherent in the mutation system. 

Phonology, in the generative use of the term, is systematic 
(morphophonological in the Prague sense). By systematic, I mean that 
the surface phonological segment (or string of segments) is related in 
a manner making the fullest possible use of generalization to an under- 
lying segment (or string of segments). Moreover, the manner in which 
the two levels are related is a dynamic one. That is, processes actively 
transform the underlying segment into a surface phonological segment. 

As we review the mutation rules offered in this chapter, we should 
be able to see that these rules fulfill the notions of systematicity 
and dynamism (through process). The surface phonological forms are 
derived from the underlying segments by processes of lenition (realized 
through various other processes), nasalization, and frication. These 
active processes are expressed in terms of phonetic feature changes. 

In order to maintain the highest degree of systematicity, it is 
necessary to strive toward the greatest amount of generalization in the 
phonological rules consistent with the organization of the phonological 
system. Herein lies a danger, if we restrict our view only to the 
phonological side without giving due consideration to the grammatical 
side of the mutation system. 

It is apparent that the process of frication is noted not onee 
but twice in the rules — once in rule 1.1, b and once in rule 1.3. At 
first glance, it would appear that rule 1.1. b and rule 1.3 should some- 
how be collapsed into a single rule in order to reflect the fact that 


they represent a single dynamic process. If we restrict our view to 
the phonological considerations of the system, such a notion that the 
rule 1.1. b in fact represents spirant mutation (becuase of the frication 
process) might appear to be justified. 

To combine rules 1,1. b and 1,3 into a single rule, however, would 
be ignoring the second half of the mutation system, the grammatical 
considerations. As the soft mutation has one set of grammatical contexts 
ana tne spirant mutation has a completely different set (as I show in 
Chapter 2), the two rules cannot be combined without creating a rule of 
such complexity as to make it unworkable. Moreover, as I demonstrate in 
section 6.1, the frication process in rule 1,1. b is actually just a 
notational device through which lenition can be realized. As stated in 
section 1.1, the soft mutation is a process of lenition that cannot 
be rendered in one single rule in binary features j rather, it must be 
rendered through a combination of other, binary features due to the 
demands placed upon generative phonology. Thus, the frication in rule 
1.1, b is just a means to another, more general process, while the frica- 
tion in rule 1.3 is the ultimate process involved. 


*The terminology used with respect to the mutation system is the 
traditional terminology used in the field of Welsh linguistics (see, 
for example, Morris Jones 1913? Jackson 1953? Watkins 1961; etc.). 
I maintain the traditional terminology in this study because it is 
clear to those working within the field, and I believe that the creation 
of new terms for the mutation system would serve no useful purpose. Of 
course, in the final analysis, the precise term used is only a con- 
venient device for designating a linguistic relationship. 

^his practice of using three distinct categories of segment 
from the underlying to the phonetic segment is established in Schane 
1973. As mentioned in section 6.2, this practice is at variance: with 
the system used in Chomsky and Halle I968 as well as in most other 
generative theoretical works. According to the system of Schane 1973 1 
the underlying segment (representation) becomes a phonological segment 
through the application of systematic phonemic rules (or morphophono- 
logical rules). The phonological segment created by these systematic 
phonemic rules then becomes a phonetic segment through the application 
of systematic phonetic rules (or allophonlc rules). Thus, the t of 
tad /tad/ 'father' becomes /d/ through the application of a systematic 
phonemic mutation rule, while the /g/ of ysgol /sgol/ 'school' becomes 
[g] (the voiceless unaspirated dorso-velar stop) through the applica- 
tion of a systematic phonetic rule. The systematic phonetic relation- 
ships are dealt with in greater detail in Chapter 7. 

3The initial ph. /f/ in the standard orthography is reserved for 
the spirant mutation of p_ /p/ only. Elsewhere the sound is rendered 
in the orthography by the letter ff . Thus, the reader, when confronted 
with an initial p_h spelled in a word should have some indication that 
the word is in the environment of the spirant mutation. 

%ote that this is the phonological position of articulation, 
not the precise phonetic point of articulation. The fact that the 
labial stops are bilabial and not labio-dental while the labial frica- 
tives are labio-dental and not bilabial indicates that the two points 
of articulation are in complementary distribution. Thus, in the phono- 
logy, we consider this to be only one position of articulation. 

5l am assuming implicitly a set of redundancy rules based upon 
the specifications in Table l.O.c. These redundancy rules apply with 
each phonological rule. Thus, for example, if a segment is specified 
[♦continuant], it need not be specified [-nasal] as in the Welsh 
phonological system, there are no segments which would be both 
[♦continuant] and [-tnasal] (no nasal fricatives). 




^This is shown in the example of phen /fen/ in note 3» above. 

7According to Mr. Hugh Jones of Bettws Garmon, Gwynedd, the 
extension of aspirate mutation is found in other districts in North 
Wales as well. 


2.0 Grammatical Motivation for Mutation . As stated In section 1.6, 
there are two sides to the mutation system — the phonological side and 
the grammatical side. In treating the latter side of mutation, it is 
necessary to consider the nature of generative phonology as a struc- 
turally based method. 

In saying that generative phonology is a structurally based system, 
I mean that the motivation for the application of any phonological rule 
is based upon notions of structural relationships. These structural 
relationships are of two types t paradigmatic and syntagraatic (compare 
de Saussure 1959). The paradigmatic relationship in the mutation 
system is the various phonological feature changes described in the rules 
in Chapter 1. These changes in effect reflect the alternation of par- 
ticular segments. 

For example, the lexical item tad |tad| 'father' can undergo rule 

1.1. a in order for dad /dad/ to be derived, or it can undergo rule 

1.2. a (l.2.c) in order for nhad /nhad/ to be derived, or it can undergo 

rule 1.3 in order for thad /Oad/ to be derived. Of course, it may not 

undergo any of these rules, in which case tad /tad/ will be derived. 

Now the segments /t/, /d/, /nh/, and /o/ alternate with one another in 


what is known as a paradigmatic relationship. As shown below, however, 
it is not a simple paradigmatic relationship. 

A paradigmatic relationship does not exist in a vaccumi it has a 



particular context, or environment. If the environment remains constant, 
then what we have in the environment is a syntagmatic relationship, or 
syntagm. For example, we can change the initial consonant of the word 
te /te/ 'tea* such that it acquires voice and loses aspiration. The 
resulting word is de /de/ 'south', which we recognize as a word with a 
different meaning from the word te /te/. On the other hand, we could 
shift the position of articulation of the initial consonant of te /te/ 
from apico-dental to bilabial, in which case we would have the word 
ne /pe/ 'if, which we also recognize as a word with a meaning quite 
different from te /te/. In these cases, only one consonant has been 
changed, and the consonants involved are in a paradigmatic relationship. 
The environment of this paradigm, however, has remained constant, -e /e/. 
This environment is a syntagm. The only things that have changed in 
this exercise are the phonological segments in initial position and the 
meaning of the resulting words. The phonological syntagm has not changed. 

The Welsh mutation system, however, represents a far more complex 
set of relationships. When an initial segment of a word changes due to 
mutation, the environment of the word itself forms the syntagm. But 
the phonological word is usually the highest point in the grammar at 
which we can talk of a syntagm for phonological items without intro- 
ducing grammatical relations. The reason why we cannot include the other 
words to which the mutated word relates in the syntagm is that the 
grammatical relationships that obtain between the affected word and the 
other words of the phrase or clause also change in the course of mutation. 

This notion of a syntagm within a grammatically well-forraed string 
of words is crucial to the operation of the mutation system. This can be 
illustrated with the above-mentioned minimal pairs, te /te/ and de /de/. 


A speaker could utter the sentence ble mae 'r te? /ble maiir te/ 'where 
is the tea? ' By changing the initial consonant of te /te/ so that it 
is voiced and unaspirated, the sentence becomes ble mae 'r de /ble maiir de/ 
•where is the south?' Of course, the sentences represent quite different 
meanings semantically, but the grammatical relationships that hold 
between the rest of the clause and te /te/ in the first sentence and 
de /de/ in the second sentence are identical. Thus, in this simple case 
of paradigmatic and syntagmatic relationships, the entire sentence 
can act as the syntagm. 

On the other hand, when a grammatical paradigm in which /t/ and /d/ 
alternate in the mutation system is set into a grammatically well -formed 
string of words, the result is quite different. To illustrate this, I 
use the example from section 1.5. In the sentence gwelodd tad /gwelo6 
tad/ 'a father saw (something)', the word tad /tad/ is the subject of 
the sentence. Now if we were to apply rule 1.1. a in order to derive 
the form dad /dad/, as we have done above, the result would be gwelodd 
dad/ 'he saw a father' , in which the word dad /dad/ is the direct object 
of the verb. Thus, the application of the mutation rules does not 
change the lexical meaning of the word, but it does reflect a change in 
grammatical relationships. 

Because the change in the phonological paradigm in the mutation 
system actually reflects a change in the grammatical relationships, we 
do not have a simple set of paradigmatic relationships existing among 
phonological items. Instead, we have two paradigms which are complexly 
interrelated — a phonological paradigm and a grammatical paradigm. 

In the transformational -generative mode of description, the 
syntactic component precedes the phonological component (see Chomsky 1965 8 


Chomsky and Halle 1968). Thus, the grammatical relationships are 
derived (transformationally) before the phonological rules are applied. 
In the dynamic and directional process of derivation, then, the particular 
item in the grammatical paradigm is established and presented to the 
(interpretive) phonological component. Because a particular grammatical 
relationship is presented to the phonological component, we apply a 
particular phonological mutation rule — the rule which corresponds to 
the grammatical relationship. It can be said, then, that these gram- 
matical relationships form the motivating environment of the mutation 
rules of Chapter 1. 

There are several pertinent considerations in the environment of 
the mutation rules. First of all, the segment to be affected must be 
in the proper position in the word—usually initial position. Further- 
more, the grammatical environment pertinent to the application of the 
rule must be present. The grammatical environment consists of two 
factors. The first of these is the morphological factor. The word 
must be of a particular class (noun, verb, etc.) and a particular gender 
and number. The second factor is the syntactic relationships into which 
the word enters (subject, object, possessed, etc.). 

Once all of the pertinent considerations for the application of 
the mutation rules are met, we can say that the motivation of the 
application of the rule lies in the environment. If the environment 
is appropriate and the segment in question is indeed covered by the 
given mutation rule, then this rule must apply. If, however, any single 
factor is absent, then the mutation rule cannot apply. Thus, each factor 
is necessary for mutation to apply, but only the sum of all the factors is 
sufficient for mutation to apply. 


The particular grammatical factors involved in mutation vary from 
dialect to dialect. As can be seen in Griffen 197^b, even the phono- 
logical factors tend to be dependent upon dialect. In Standard Welsh, 
however, they are fairly consistent. A complete description of them can 
be found in Morgan 1952 along with pertinent exceptions (counter-examples 
in the sense of Hjelmslev 1970:30-1) and the historical development 
of various forms in Modern Welsh. Short but usually adequate summations 
are often given in textbooks (such as Bowen and Rhys Jones 19^0; James 
1966 j etc.). 

In the following sections, I describe the grammatical environments 
of the various mutation rules. This is only intended to be a short 
summation in generative terminology, I use the classifications and order 
of James (L 966 141 -3) because of its accessibility, organization, and 

2.1 Environment I . Before treating those environments in which a 
mutation rule does apply, I should first clarify what takes place when 
the environment of a word does not in fact satisfy the conditions of 
the mutation rules. When the environment does not meet any of the 
mutation rules, the underlying segment (the radical) is realized as 
the phonological segment. Whenever the environment does not specify 
the application of any mutation, it is termed environment I, To specify 
the exact description of environment I, as well as the exact description 
of environments which cause the mutation rules to apply, would be very 
redundant in the description (compare the practice of Gleason I96I J 


For example, the personal possessive pronouns motivate the environment 


for all mutations in the system. The first person singular possessive 
pronoun motivates the nasal mutation, the second person singular 
motivates the soft mutation, the third person masculine also motivates 
the soft mutation, the third person feminine motivates the spirant 
mutation and aspirate mutation, and the first and third person plural 
motivate the aspirate mutation. Now as it happens, the second person 
plural possessive pronoun has not been specified as being a motivating 
factor for any of the mutation rules. By its complementary nature, 
we can therefore assume that environment I obtains and that none of the 
mutations apply. Thus, the radical is derived as the phonological seg- 
ment, with no feature changes. So eich tad /eix tad/ 'your (plural) 
father' does not reflect the application of any mutation rule. 

This notion of environment I may appear to be fairly uninteresting 
at this point, but it proves to be quite crucial. The point that must 
be remembered is that when no mutation environment exists, the radical 
is realized. Moreover, this radical, as shown in section 1,0, is to be 
taken as the underlying form, the segment from which all of the mutation 
forms are derived. 

At first glance, there does appear to be a complication in our 
handling of the radical as the underlying segment and of environment I 
as the general environment. This complication centers around the 
notion of 'hard mutation' (see Watkins I96I166-7). According to the 
traditional analysis of the system, hard mutation affects adjectives in 
such a way that an adjective ending in a voiced unaspirated stop 
'hardens' the stop to a voiceless aspirated one when an ending (equative, 
comparative, or superlative) is added. 

Hard mutation appears to have all of the elements of a normal 


consonant mutation — a particular phonological class (mediae) and 
grammatical considerations (the class of adjectives and particular 
endings). For example, the adjective teg /teg/ 'fair' is realized as 
teced /teked/ in the equative, tecach /tekax/ in the comparative, and 
tecaf /tekav/ in the superlative. Likewise, tlawd /tlaud/ 'poor' is 
realized as tloted /tloted/ in the equative, tlotach /tlotax/ in the 
comparative, and tlotaf /tlotav/ in the superlative. Moreover, gwlyb 
/gwlib/ 'wet' is realized as gwlyped /gwleped/ in the equative, gwlypach 
/gwlepax/ in the comparative, and gwlypaf /gwlepav/ in the superlative. 
Thus, the entire natural class of voiced unasplrated stops is affected. 

If we were to accept the traditional notion of hard mutation, 
then we would have to state that in this one case environment I motivates 
the application of a rule by which the underlying voiced unasplrated 
stops become realized as voiceless and aspirated. For several reasons, 
however, I do not accept the traditional notion of hard mutation. First 
of all, while the soft mutation is quite extensive in its range of 
application, the hard mutation is restricted to this one type of instance, 
by treating the voiceless aspirated stop as underlying and extending 
the environment of the soft mutation rules to include segments in final 
positlon at least in adjectives, we could simplify the system of rules 
and simplify our description. Furthermore, as I show in section 
5.1, there has been a general historical development whereby the final 
stop in the Welsh word has become voiced and unasplrated (although, t6 
be sure, voicing in final position is not frequent in languages). Taking 
this development into account, the soft mutation would normally apply 
to the final position, anyway, making the application of the soft muta- 
tion rule 1.1. a in this case simply a reflection of a much larger 


phenomenon. This analysis does not reflect a grammatical environment, 
and this problem is also discussed in section 5.1. 

One argument that could be used for the traditional interpretation 
of the hard mutation is the fact that in Middle Welsh, an /h/ sometimes 
preceded the adjective ending (see Evans 1960:23-4-). This /h/, however, 
is no longer productive. Although it is an important consideration in 
an earlier phonology of Welsh, it is not pertinent to Modern Welsh — 
especially not in a synchronic description of Modern Welsh, I return 
to this relationship between /h/ and the stops in Chapter 7, 

2.2 Environment II , Listed below are the structural descriptions of 
most environments which, in Standard Welsh, are sufficient for the 
application of the soft mutation rules. Unless otherwise stated, it 
is understood that the segment in the critical position for mutation is 
in fact any one of the affectable segments listed in section 1,1, For 
each structural description, I first discuss the necessary elements 
and then form these into the generative model (see, for example, Chomsky 
I9651 Jacobs and Rosenbaum 1968j R.M, Jones 1963). All descriptions 
are based upon James (I966«4l-3) f and full descriptions are found in 
Morgan 1952, Because of the large number of environments found in 
environment II, it is usually necessary to list first those which affect 
nouns (environments Ila-m), then those which affect adjectives (environ- 
ments Iln-s), these which affect verbs (environments Ilt-z), and finally 
that which affects adverbs (environment Ilaa), In any case, the word 
affected by soft mutation is marked in the formal structural description 
with the notation [+LM] (for lenis mutation). 

a. Welsh has only one article—the definite article. This is 


realized as y_ /a/ before words beginning with consonants, vr /or/ 
before words beginning with vowels (including the case in which I g I is 
deleted through soft mutation), and 'r /r/ after words ending in a 
vowel (regardless of the following word). When any one of these forms 
of the definite article precedes a noun which is both feminine in 
gender and singular in number, the initial consonant of the noun is in 
environment II. However, the environment Ha is only operative for the 
mutation rules l.l.a-d—rule 1.1. e (11 /l/ and rh /»/) does not apply 
in this environment. For example, dafad /davad/ 'sheep' is a singular 
feminine noun. When preceded by the definite article y_ /a/ , the noun 
phrase is realized as y_ ddafad /o 6avad/ 'the sheep'. The rule does 
not apply, however in the case of Haw /lau/ 'hand', which is realized 
£ Haw /q lau/ 'the hand'. 

The formal structural description for environment Ha can be 
rendered as follows! 



' 4f eminine " 



-rule 1.1, e 

b. The second person singular familiar possessive pronoun is 
realized as dy_ /do/ 'thy' or as 'th /©/ when it is contracted onto the 
preceding word ending in a vowel (the actual words with which the 
pronouns can be contracted are fairly restricted in number). The third 
person singular masculine possessive pronoun is realized as ei /ii/ 
•his' or as the contracted forms 'i /i/ or 'w /u/ (the latter when the 
preceding word ends in /i/ or /i/). These pronouns always precede 


a verb-noun, ^ which in this case operates in the mutation system as a 
noun, as the direct object of the verb-noun. The noun or verb-noun is 
thus in environment lib, and the initial consonant undergoes any applic- 
able soft mutation rule. For example, tad /tad/ 'father' can be realized 
as dy_ dad /da dad/ 'thy father' or ei dad /it dad/ 'his father'. The 
periphrasic verbal phrase for *I see him' would be wyf i^ vn ei weld 
/u«iv 1 en i« weld/ with the verb-noun derived from gweld /gweld/ 'to 

The formal structural description for environment lib can be 
rendered as follows « 

(verb-)noun phrase 




-feminine _ 


c. Twelve of the more common prepositions are am /am/ 'for', ar 
/ar/ 'on', at /at/ 'to', gan /gan/ Vith', tros /tros/ 'over', trwy 
/tru»i/ 'through', wrth /ure/ 'at', dan /dan/ 'under', heb /heb/ 
'without', hyd /hid/ 'till', i /i/ 'to', and o /o/ 'from'. These 
constitute an element of environment He. If a noun follows any one of 
these, the initial consonant of the noun will undergo soft mutation. 
This group of prepositions, moreover, with two exceptions represents 
the class of prepositions which are conjugated.^ For example, the 
preposition ar /ar/ can precede the noun cae /caii/ 'field' for the 
prepositional phrase ar gae /ar gaii/ 'on a field'. 

By marking the exceptions in the lexicon, we can render the above 
environment as follows « 




[-taonjugatable] [+LM] 

d. A noun phrase can be constructed, with a numeral and a noun. 
The noun in this case is always singular. The noun is in environment 
lid, if the numeral is un /in/ 'one' (which precedes both masculine and 
feminine words) and the noun is feminine in gender or the numeral is 
either dau /dai/ 'two (masculine)' or dwy /duii/ 'two (feminine)'. 
For example, merch /raerx/ 'girl' is feminine, 'One girl' would be 
un ferx /in verx/, and 'two girls' would be dwy ferch /duii verx/. 

The formal structural description for environment lid can be 
rendered as follows* 



One "A 
Two ) 

(f""0ne 1) [+LM] 


I Two J 

e. Usually, an adjective follows the noun it modifies. Some 
adjectives, however, regularly precede the noun, and for special 
emphasis any adjective can precede the noun it modifies. When an 
adjective does precede the noun, the noun is in environment lie, and 
the initial segment is subject to soft mutation. For example, dyn 
/din/ 'man' usually follows the adjective hen /hen/ 'old', resulting 
in hen ddyn /hen 6 in/ 'old man'. Compare mab unig /mab inig/ 'lonely 
son ( mab )' with unig fab /inig vab/ 'only son'. 

The formal structural description for environment lie can be 
rendered as follows t 




f . A predicate noun In Welsh always follows the predicator yn 
/en/. When this occurs, the noun (but not a verb-noun) is in environ- 
ment Ilf , and its initial consonant undergoes soft nutation. For 
example, Cymro /kemro/ •Welshman' undergoes rule 1,1. a in the sentence 
mae ef yn Gymro /mati ev an gararo/ 'he is a Welshman', As in environment 
Ila, environment Ilf does not effect rule l.l.e. 

The formal structural description for environment Ilf can be 
rendered as follows i 


[-rule 1.1. ej 

g. In Welsh, either an inflected form of the verb may be used or 
a periphrastic verbal phrase (verb 'to be' + predicator + verb-noun). 
When the inflected form of the verb is used, any noun which serves as 
direct object to the verb is in the environment Ilg, and its initial 
consonant undergoes soft mutation. For example, wyf i yn gweld dyn 
/utiV i en gweld din/ 'I am seeing a man ( dyn )' can be rendered with an 
inflected verb form as gwelaf ddyn /gwelav 6 in/. 

The formal structural description for environment Ilg can be 
rendered as follows t 


verb noun phrase* noun phrase 2 


This notation is in accord with R.M. Jones 19&3. The second 
noun phrase in this system is the direct object. 

h. Normally, the noun subject follows the verb 'to be' in a 
periphrastic verbal construction or with an Inflected verb, and the noun 
object follows the verb-noun in a periphrastic verbal construction (see 
the previous section). Whenever this pattern is interrupted by an 
adverbial element (for example, a sentential adverb or prepositional 
phrase), the noun affected is in environment Ilh, and its initial con- 
sonant undergoes soft mutation. For example, the sentence 'there were 
people ( pobl ) there (yno ) on vacation' can be rendered without interrup- 
tion as oedd pobl yn gwylio yno /oi46 pobl on guiilio ano/ or with 
interruption oedd yno bobl yn gwylio /oti& ©no bobl on gu»ilio/. 

The formal structural description for environment Ilh can be 
rendered as follows t 



In the case of periphrastic verbal constructions (for example, oedd 
yno bobl yn dawns lo /oe6 ©no bobl en daunsio/ 'there were people there 
dancing'), following R.M. Jones 1963. the following structural descrip- 
tion, derived from a transformation, would apply: 




* Bod "I 
llTo be J 




i. When a title or a term in apposition follows a noun, the title 
is in environment Hi, and the initial consonant undergoes soft mutation. 
For example, in the phrase Llewelyn Frenin /leuelin vrenin/ 'Llewelyn 
the King', brenin /brenin/ 'king* undergoes soft mutation because it 
is used as a title. In the phrase Llewelyn Fawr /leuelin vaur/ 
'Llewelyn the Great', mawr /maur/ undergoes soft mutation even though 
it is an adjective normally, because it is functioning as a noun in 
apposition in this phrase. (Compare Llewelyn mawr /leuelin maur/ 
•great Llewelyn', in which it is used as an adjective.) 

The formal structural description for environment Hi can be 
rendered as follows* 



+LM J 

j. The vocative in Welsh is marked such that a noun used in a 
greeting is in environment II j, and its initial consonant undergoes 
soft mutation. For example, the greeting '0 God' is rendered as 
Dduw /o 6iu/, a soft mutation form of Duw /diu/ 'God'. 

The formal structural description for environment IIj can be 
rendered as follows t 



+LM J 

k. Compound words are often formed by joining two nouns into a 
single noun. When this is done, the second noun-morpheme is in environ- 
ment Ilk, and its initial consonant undergoes soft mutation. (Compare 
this with environment Hi, above.) For example, the noun gwalth /gwaie/ 
•work' joins with the noun t£ /til/ •house' in order to produce the 
noun gweithdy /gwei©di/ 'workshop'. 

The formal structural description for environment Ilk can be 
rendered as follows: 


1. When the conjunction neu /nei/ 'or* joins two nouns, the noun 
following the conjunction is in environment III, and its initial con- 
sonant undergoes soft mutation. For example, bachgen /baxgen/ 'boy* 
and merch /merx/ 'girl' can be joined as bachgen neu ferch /baxgen nai 
verx/ 'a boy or a girl'. 

The formal structural description for environment III can be 
rendered as follows t 


[Neu] C+LM] 

m. In Welsh, there are a number of deictic words which function 


as nonconjugatable verbs. The most common of these are dyraa /dema/ 
•here is', dyna /dena/ 'there is', and dacw /daku/ 'yonder is'. They 
can be used either in the singular or in the plural. When one of these 
deictic verbals is used, the noun to which it refers is in the environ- 
ment lira, and its initial consonant undergoes soft mutation. For example, 
Hong /lop/ 'ship' can be the object of dacw /daku/ to yield the sen- 
tence dacw long /daku log/ 'yonder is a ship'. 

The formal structural description for environment lira can be 
rendered as follows i 


[•hleictic] [+LM] 

n. As stated in reference to environment He, the adjective 
usually follows the noun it modifies. If the noun is both feminine 
in gender and singular in number, then the adjective following the 
noun is in the environment Hn, and its initial consonant undergoes 
soft mutation. For example, when da /da/ 'good' follows bachgen /baxgen/ 
•boy' the result is bachgen da /baxgen da/ 'a good boy', but when it 
follows merch /merx/ 'girl' the result is merch dda /merx 6a/ 'a good 

The formal structural description for environment Iln can be 
rendered as follows i 



[^singular] [+LM] 



o. An adjective can be used as an adverb by placing it after the 
predicative yn /en/. When this is done, the adjective is in the environ- 
ment IIo, and its initial consonant undergoes soft mutation. For example, 
the adjective da /da/ means 'good', but the phrase yjn dda /en 6a/ 
means 'well'. 

The formal structural description for environment IIo can be 
rendered as follows t 




p. The comparison of adjectives in the equatlve (as... as) is 
accomplished in Welsh by placing mor /mor/ before the adjective and 
& /ai/ after it or by placing cvji /kin/ before the adjective and a 
/ai/ after it. The adjective following either mor /mor/ or cyjn /kin/ 
is in the environment Up, and its initial consonant undergoes soft 
mutation. The comparison of adjectives in the comparative, moreover, 
is accomplished by placing jn /on/ before the adjective and na /na/ 
after it. The adjective following vn /an/ is also in environment Up, 
and its initial consonant undergoes soft mutation. For example, gwyn 
/gwin/ •white' undergoes rule 1.1. c in the phrase mor wyn a'r eira 
/mor win a:r eira/ *as white as snow'. However, when the adjective is 
preceded by cyn, rule 1.1. e (11 /i/ and rh /*/) is not effected. 

The formal structural description for environment Up can be 
rendered as follows* 




if Mor V L<-rule 1 . 1 . e>J lit 

The angled brackets (< >)can be read 'only if cyn /kin/ is used, the 
rule will not effect rule l.l.e'. 

q. There are certain adverbs that often precede and modify adjec- 
tives. These include rhy /»i/ •too', pur /p4r/ 'pure', lied /led/ 
•almost', gweddol /gweool/ 'fairly*, go /go/ 'quite', and hollol 
/holol/ 'completely'. Whenever one of these adverbs precedes an adjec- 
tive, the adjective is in environment Hq, and its initial consonant 
undergoes soft mutation. For example, tenau /tenai/ 'thin' can be 
preceded by rhy /*i/» and tne resulting phrase is rhy denau /a?i denai/ 
•too thin'. 

These adverbs can be marked in the lexicon, and the formal struc- 
tural description for environment Ilq rendered as follows* 




r. As we see in environment Ilg, when a noun is the direct object 
of an Inflected verb, its initial consonant undergoes soft mutation. 
If an adjective precedes the noun, it is in environment Ilr, and its 
initial consonant undergoes soft mutation. ^ For example, the periphras- 
tic construction wvf i wedi clywed rhyw tjr /uxiv i wedi kliued ariu 
u»r/ 'I heard some ( rhyw ) man' can be rendered with the inflected verb 


as clywais ryw wr /kliuais riu utr/. 

The formal structural description for environment Ilr can be 
rendered as follows i 



1 [+LH] 2 

s. In environment III, we see that when two nouns are conjoined 
by neu /nei/ 'or', the noun following the conjunction is in environ- 
ment III. When two adjectives are conjoined by neu /nai/, the second 
adjective is in environment lis, and its initial consonant undergoes 
soft mutation. For example, when gwyn /gwin/ 'white' is conjoined with 
du /di/ 'black' by neu /nai/, the result is gwyn neuddu /gwin nai 61/ 
'white or black'. 

The formal structural description for environment lis can be 
rendered as follows i 


[Neu] [+LM] 

t. When the inflected form of the verb is used in a relative 
clause, the relative pronoun (either subject or object) is realized 
as a /a/ in the affirmative and na /na/ in the negative. The inflected 
form of the verb must follow the relative pronoun, and the verb is in 
the environment lit, its initial consonant undergoing soft mutation. 
For example, we find the phrase y_ bachgen na welodd y_ ffalr /a baxgen 
na weloo a fair/ 'the boy who did not see ( gwelodd ) the fair'. When 

na /na/, the negative relative pronoun, is used, this rule does not 
effect rule 1.1. a (p_ /p/, t /t/, c /k/) (see environment IVf), 

The formal structural description for environment lit can be 
rendered as follows t 





<-+negative> +LM 

L<-rule l.l.a>J 


u. The conjunction pan /pan/ 'when' always precedes the verb. 
When it is used, the verb following pan /pan/ is in the environment 
IIu, and its initial consonant undergoes soft mutation. For example, 
there is a phrase pan ddeuthum /pan 6ai©im/ 'when I came ( deuthum ) ' . 

The formal structural description for environment IIu can be 
rendered as follows! 





v. There are several sentential affirmative particles that may 
optionally introduce the sentence. These precede the verb and include 
fe /ve/, mi /mi/, and ti /ti/. When these are used, the verb is in 
environment IIv, and its initial consonant undergoes soft mutation. 
For example, the sentence cefais i afal /kevais i aval/ *I got an apple' 
can alternatively be realized as fe gefals i afal /ve gevais i aval/. 

The formal structural description for environment IIv can be 


rendered as follows i 




w. The usual way of asking a question is by placing the inter- 
rogative a /a/ in front of the sentence. In this position, it precedes 
the verb, which is in environment IIw, and the initial consonant of the 
verb undergoes soft mutation. For example, the sentence llwyddodd ef 
/lu»i6o6 ev/ 'he succeeded' can be turned into a question with a /a/, 
a lwyddodd ef? /a lu»i6o6 ev/ 'did he succeed?' The environment IIw 
is not sufficient to effect rule 1,1. a (p_ /p/, t /t/, c /k/). 

The formal structural description for environment IIw can be 
rendered as follows t 

[A] «H 

U-rule 1.1. aj 


x. Sentences are rendered in the negative by placing a negative 
word at the beginning of the sentence, such as ni /ni/, na /na/, and 
oni /oni/ (negative interrogative). The verb which follows such a 
negative is in environment IIx, and its initial consonant undergoes 
soft mutation. For example, daeth %_ bws /da tie a bus/ 'the bus came' is 
negated as ni ddaeth £ bws /ni 6ai49 a bus/ 'the bus did not come'. 
Environment IIx is not sufficient to effect rule 1.1. a (2 /p/, t /t/, 
c /k/) (see environment IVd). 


The formal structural description for environment IIx can be 
rendered as follows: 



Urule 1.1. a] 

y. An indirect question is introduced by the particle a /a/. 
The verb form following the particle is in environment Ily, and its 
initial consonant undergoes soft mutation. For example, we find the 
sentence gofyn a gyrhaeddodd hi /govin a gerhaiiooo hi/ *he asks whether 
she has arrived ( cyrhaeddodd )'. 

The formal structural description for environment Ily can be 
rendered as follows: 


[ -Unterrogat ive] 

[A] [+LM] 


z. The contracted form of the possessive pronoun can be infixed 
between the affirmative sentential particle (see environment IIv) and 
the verb. As such, it functions as the pronominal direct object. When 
this pronoun is that of the second person singular (familiar), the following 
form of the verb is in environment Hz, and its initial consonant under- 
goes soft mutation. For example, there is the sentence fe'th welais 
/ve© welais/ 'I saw ( gwelais ) thee'. 

The formal structural description for environment Hz can be 


rendered as follows t 



[+singularl [+LM] 
m +second J 

aa. Any sentential adverb or adverbial phrase is in environment 
Ilaa, and its initial consonant undergoes soft mutation. This applies 
no matter where in the sentence it may appear. For example, we have 
the sentence gwelais ef ddoe /gwelais ev 6o«i/ 'I saw him yesterday 
(doe) ' . 

The formal structural description for environment Ilaa can be 
rendered as follows i 



2.3 Environment III . The following is a list of the structural descrip- 
tions of most environments which are sufficient for the application of 
the nasal mutation rules in Standard Welsh, As with the soft mutation, 
I adhere to the list and order found in James (l 9661 41 -3), the full 
descriptions for which are found in Morgan 1952. The word affected by 
nasal mutation is marked in the formal structural description with the 
notation [-WM], 

a. The first person singular possessive pronoun in Standard Welsh 
is fy_ /vo/ 'my. The noun which follows this pronoun is in the environ- 
ment Ilia, and its initial consonant undergoes nasal mutation. For 


example, cartref /kartrev/ 'home' when possessed by the first person 

singular pronoun is realized as fy_ nghartref /ve ghartrev/ 'my home'. 


The formal structural description for environment Ilia can be 
rendered as follows! 



Ppossesslve"| [-WM] 

L +first J 

As in environment IIb t the pronoun can also precede the verb-noun in 
the periphrastic verbal construction, as in the sentence mae ef yn fy 
ngweld /ma si ev an ve nweld/ 'he sees ( gweld ) me'. In this case, the 
first person singular possessive pronoun functions as the direct object, 

b. In Standard Welsh, there is a preposition vn /en/ 'in'. When 
this preposition precedes a noun, the noun is in environment Illb, and 
its initial consonant undergoes nasal mutation. For example, 'Wales' 
is Gymru /kemri/, but 'in Wales' is yng Nghymru /en phamri/. As can be 


seen from this example, the preposition is actually /eN/, where /n/ is 
an archiphoneme in which position of articulation is predictable and 
takes on the value of the following consonant or /n/ (see Griff en 197 / to). 

The formal structural description for environment Illb can be 
rendered as follows: 



[Yn] [-WM] 

c. In reckoning time, there are three words of particular importance — 


blwydd /blu»46/ 'year (of age)', blynedd /blaneo/ 'year', and diwrnod 
/diurnod/ 'day'. As stated in section 2.2.d, when a noun follows a numeral 
it is realized in the singular. When any of these three nouns follows 
pum /pirn/ 'five', saith /saiO/ 'seven', wyth /uiie/ 'eight', naw /nau/ 
•nine', deg /deg/ 'ten', deuddeg /daioeg/ 'twelve', pymtheg /pamOeg/ 
'fifteen', deunaw /dainau/ 'eighteen*, ugain /igain/ 'twenty', or can 
/kan/ 'hundred', it is in the environment lie and its initial consonant 
undergoes nasal mutation. For example, 'seven years' is saith mlynedd 
/sal© mlaneo/, and 'eight days' is wyth nlwrnod /u»49 niurnod/. 

The formal structural description for environment IIIc can be 
rendered as follows « 



Pum *\ fBlwydd ") 

Saith J \ Blynedd { 

Wyth / (Diwrnod ) 

Naw / L -WM 

Deg I 

Deuddeg f 

Pymtheg I 

Deunaw I 

Ugain I 

Can J 

Those numbers not listed in the description either cause no mutation or 
cause soft mutation (enironment lid) or spirant mutation (environment IVa). 

d. There is a morpheme prefix vn- /an/, not to be confused with 
the preposition of the same phonological shape, which has various 
meanings depending upon the word to which it is prefixed. When the 
prefix is unstressed, the word to which it is prefixed is in environment 
Hid, and its initial consonant undergoes nasal mutation. Examples 


are found in ynghylch /snhilx/ 'about* (from jn /an/ plus cylch /kilx/ 


'circle' )» ymble /emhle/ 'in what place' (from vn /en/ plus pie /pie/ 

•place'), and ynghyd /enhid/ 'together' (from vn /en/ plus /k4d/ 



The formal structural description for environment lid can be 
rendered as follows t 


r In 1 [■««] 

2.4 Environment IV . Below are the structural descriptions of most 
environments which are sufficient for the application of the spirant 
mutation rule in Standard Welsh, Again, I adhere to the list and order 
found in James (l966»4l-3), the full descriptions for which are in 
Morgan 1952. The word affected by spirant mutation is marked in the 
formal structural description with the notation [+SM], 

a. As in sections 2,2.d (environment lid) and 2.3.c (environment 
IIIc), when a numeral precedes a noun, the noun is realized in the 
singular. Now if the numeral preceding the noun is either tri /tri/ 
•three' or chwe /xwe/ 'six', then the noun following the numeral is in 
environment IVa, and its initial consonant undergoes spirant mutation. 
For example, when the word ceffyl /kef 41/ 'horse' is preceded by chwe 
/xwe/, the resulting phrase is chwe cheffyl /xwe xefil/ 'six horses'. 

The formal structural description for environment IVa can be 
rendered as follows! 





b. The third person singular feminine possessive pronoun is ei 
/it/ 'her*. It can also be realized as the contracted forms 'i /i/ 
and 'w /u/, the latter when the preceding word ends in /i/ or /i/. 
As in section 2,2.b (environment lib), the contracted forms, identical 
with the masculine (as is the neutral form), can be added to the end 
of a few words ending in vowels. When any form of the third person 
singular feminine possessive pronoun precedes a noun, the noun is in 
environment IVb, and its initial consonant undergoes spirant mutation. 
As with all possessive pronouns, these can precede the verb-noun of a 
periphrastic construction and operate as the direct object of the 
verb-noun, in which case the verb-noun is also in environment IVb, 
For example, pen /pen/ means 'head', and ei phen /ii fen/ is 'her head". 
Likewise, we find the sentence wyf .1 wedi ei chlywed /uiiv i wedi ii 
xliued/ 'I heard (clvwed) her*. 

The formal structural description for environment IVa can be 
rendered as follows i 






c. Three prepositions often found in Standard Welsh are a /an/ 
•with', gyda /gsda/ 'along with', and tua /tia/ ' towards' . Whenever 


one of these prepositions precedes a noun, the noun is in environment 
IVc, and its Initial consonant undergoes spirant mutation. For example, 
we find the phrase gyda ohyfeillion /geda xavsilion/ 'along with 
friends ( cyfeilllon )'. 

The formal structural description for environment IVc can be 
rendered as follows i 




| Gyda L 
I TuaJ 

d. It was pointed out in section 2.2.x (environment IIx) that 
sentences are negated by preceding them with negative markers including 
ni /ni/, na /na/, and oni /oni/ (negative interrogative). Although 
environment IIx does not affect the voiceless aspirated stops, environment 
IVd does. Thus, the verb form following a negative in the sentence is 
in the environment IVd, and its initial consonant undergoes spirant 
mutation. For example, there is the question oni thai odd hi? /oni 
9alo6 hi/ 'did she not pay ( talodd )? ' I address the peculiarities of 
the environments as regards environment II and environment IV in 
section 2.6, 

The formal structural description for environment IVd can be 
rendered as follows i 



e. As pointed out in section 2.2.q (environment Hq.), some adverbs 
marked in the lexicon cause a following modified adjective to undergo 
soft mutation. In the case of tra /tra/ 'very', however, the following 
adjective is in environment IVe, and its initial consonant undergoes 
spirant mutation. For example, caredig /karedlg/ 'dear' can be preceded 
by tra /tra/ to yield the phrase tra charedig /tra xaredig/ 'very dear*,' 

The formal structural description for environment IVe can be 
rendered as follows: 



[Tra] [+SM] 

f. The three most common sentence conjunctions are a /a/ 'and', 
na /na/ 'nor', and oni /oni/ 'until' (not to be confused with the 
negative interrogative). Any word following one of these conjunctions, 
whether noun or verb, is in environment IVf , and its initial consonant 
undergoes spirant mutation. For example, we find ci a chath /ci a xaO/ 
•a dog and a cat ( cath )' , and oni chyrhaeddodd v_ tren /oni xsrha»i6o6 
9 trein/ 'until (unless) he caught ( cyrhaeddodd ) the train'. 

The formal structural description for environment IVf can be 
rendered as follows: 




g. As shown in section 2.2.t, the negative form of the relative 


pronoun na /na/ is not sufficient to effect rule 1.1. a in environment 
lit. However, when the negative relative pronoun precedes a form of 
the verb, the verb is in environment IVg, and its initial consonant 
undergoes spirant mutation. For example, there is the sentence dyma 'r 
gwr na chafodd arlan /da mar guir na xavoS arian/ 'here is the man who 
did not get ( cafodd ) money'. As in section 2.4.d, I return to the 
matter of the apparent conflict in environments in section 2.6. 

The formal structural description for environment IVg can be 
rendered as follows: 



[+negative] [+SM] 

Before leaving the spirant mutation (environment IV), I should 
like to address the matter brought out in section 1.6. Now that the 
environments for the soft mutation and the spirant mutation have been 
examined, it should be fairly obvious why we cannot combine the 
Ibl - /v/ and Idl ■• /6/ rule (l.l.b) with rule 1.3, in spite of the 
fact that /v/ and /6/ are, after all, spirants, or fricatives. If we 
were to combine rules l.l.b and 1,3» we would have to mark twenty -five 
environments (considering that two overlap) so that part of the spirant 
mutation rule could apply and part could not, and we would have to mark 
five more environments so that the latter part could apply and the former 
part could not. This would make it virtually impossible to combine the 
two phonological rules. 


2.5 Environment V , The final environment that concerns us is the 
environment which motivates rule 1,4, the aspirate mutation. As 
pointed out in section 1,4, the aspirate mutation is not exactly a 
part of the consonant mutation system, as it does not affect consonants 
(at least in Standard Welsh), though it does affect initial position. 
Nonetheless, the same considerations affect the environment of aspirate 
mutation as affect the environments of the various consonant mutations. 

Although James 1966 does not list aspirate mutation among the 
consonant mutations, Bowen and Rhys Jones (I96O1I67) does list it, 
and Morgan 1952 also treats it. The word affected by aspirate mutation 
in the formal structural description is marked [+AM], 

The one environment involved in aspirate mutation has to do with 
the possessive pronouns. When the noun or the verb-noun (in the case of 
periphrastic verbal constructions) follows the first person singular 
possessive pronoun contraction 'm /m/ 'my 1 , the third person singular 
feminine possessive pronoun ei /it/ or its contractions 'i /i/ or 
'w /u/ 'her', the first person plural possessive pronoun ein /a in/ 
or its contraction 'n /n/ 'our', or the third person plural possessive 
pronoun eu /ai/ or its contractions 'u /i/ or 'w /u/ 'their', then 
the noun is in environment V, and its initial vowel undergoes aspirate 
mutation. For example, 'name' is enw /enu/, and 'her name* is ei henw 
/i» henu/. 

The formal structural description for environment V can be 
rendered as follows: 




!' +singular " 
" +f irst "I 
'♦third T 
+femininej i 
: +plural = . 
. -second m J 



2.6 Overlapping Environments . In section 2.2.t we find that the 
negative relative pronoun is not sufficient to effect rule 1.1. a, and 
in section 2,2,x we find that the same is true of all negative sentence 
markers. By using notation, we can describe the phenomenon in which 
rule 1.1. a is excluded from the environments while rules 1.1. b-e are 

As we shift our attention to sections 2,**.d and 2,4,g, we wee that 
the spirant mutation is effected in precisely those environments in 
which the negative relative pronoun and negative sentence markers are 
found. Thus, it would appear that environments lit and IIx overlap 
with environments IVg and IVd. That is to say that the grammatical 
specifications are identical. 

The notational device which we can use in order to determine when 
one mutation applies and when the other applies is known as rule 
ordering. In the generative approach to grammar, the surface (real) 
level is derived by rules from the deep (abstract) level. Each rule 
application creates its own intermediate stage with its own item, upon 
which further processes work in order to derive the ultimate surface 
forms. These rules must, then, apply in a particular order, and the 
order used is determined by the generality of one particular order over 


another (for examples of such rule ordering, see S. Anderson 1969; 
Koutsoudas, Sanders, and Noll 1974), 

If, once all environments are established, we order the soft 
mutation rules first, then, in accordance with the [-rule 1.1. a] 
notation in environments lit and IIx, we must make sure that the soft 
mutation rule 1.1. a is not effected. So long as the voiceless aspirated 
stops do not undergo rule 1.1. a in environments lit or IIx, then they 
will be present for the application of rule 1.3 in environments IVg or IVd. 

On the other hand, if we were to order the spirant mutation rule 
before the soft mutation rules, then the spirant mutation rule 1.3 
would eliminate all occurrences of the voiceless aspirated stops in 
environments IVg and IVd. Once the voiceless aspirated stops in environ- 
ments IVg and IVd are eliminated, there is no reason for specifying in 
the environments lit and IIx that the soft mutation rule 1.1. a is not 
effected, for there are no longer any segments which the rule can 
affect. Thus, we can allow rule 1.1. a to apply, as it applies vacuously. 

For example, we can consider the two sentences dyma 'r dyn na 
chlywals /demar din na xliuais/ 'here is the man whom I did not hear' 
and dyma 'r dyn na welals /demar din na welais/ 'here is the man whom I 
did not see'. The verb in the first relative clause is clywais /kliwais/, 
and that in the second is gwelals /gwelais/. As it stands now, environ- 
ments lit and IIx are specified in such a way as to prohibit the underlying 
|k| from undergoing rule 1.1. a and deriving an inadmissable */gliiaa.!Ls/ , 
If, however, we apply the spirant mutation first, the underlying I kl is 
rewritten /x/, to yield the correct form /xliuais/. If the spirant 
mutation applies first, nothing happens to the underlying |g| of 
/gwelais/, as rule 1.3 only applies to voiceless aspirated stops. If 


we apply the soft mutation rules now, there is no /k/ upon which rule 
1,1, a can act, and so there is no reason to specify that 1.1. a does not 
apply. Thus, the spirant mutation 'bleeds* (see Kiparsky 1968) the 
environment of the soft mutation rules. 

In this manner, the overlapping of environments can lead to a more 
economical description. By taking the bleeding relationship into 
account, we can reduce the specification in environments lit and IIx, 
yielding simpler rules. 

Moreover, we now find that environment lit is identical to environ- 
ment IVg and environment IIx is identical to environment IVd, except 
for the marking of the particular mutation. Thus, the two pairs can be 
combined with the mutation markings in braces. Of course, [+SM] must 
be entered above [+LM], as spirant mutation must precede soft mutation. 
We can refer, then, to such overlapping environments withthe notation 
'environment IV/lI'. 

2.7 Constituent Environments . A close examination of the environments 
put forth in this chapter should reveal some potential conflicts. For 
example, in environment IVa, we see that a noun possessed by the third 
person singular feminine possessive pronoun is in a spirant mutation 
environment. Moreover, in environment Ilg, we see that a noun which 
functions as the direct object of an inflected verb is in a soft muta- 
tion environment. What happens if the noun functioning as the direct 
object of an inflected verb (environment Ilg) happens to be preceded 
by the third person singular possessive pronoun (environment IVa)? 

Let us take, for example, the word cath /kaO/ 'cat' with an 
underlying initial |k|. The initial segment undergoes rule 1.1, a in 


the sentence gwelals gath /gwelais ga©/ 'I saw a cat', and it undergoes 
role 1.3 in the phrase ei chath /ij xa©/ 'her cat'. When the phrase is 
embedded within the sentence, the result is gwelais ei chath /gwelais 
it xa©/ 'I saw her cat*, in which the spirant mutation prevails. This 
is not a situation, as we find in the previous section, in which the 
more restricted environment applies first, for we can also find the phrase 
ei hen gath /ii hen ga©/ 'her old cat', in which environments He and 
IVa conflict and environment He prevails. 

The choice of which environment prevails is dependent upon the 
immediate constituents of the sentence (as adapted from Wells 19^7 1 
Postal 196^ j etc.). An immediate constituent analysis of the above 
example would yield the following tree diagram » 






Under NOUN PHRASE 2 , NOUN is an immediate constituent of DETERMINER, and 
the entire NOUN PHRASE2 is an immediate contituent of VERB under 

Thus, we can form a generalization as to the application of the 
various environments. A word can only be affected by the environment 
of its immediate constituency. This forms an important element in the 
description of the mutation system, for it limits the effective range 
of dominance of the mutation rules. 

An interesting example of this phenomenon is found in section 


2,2,r, In the sentence clywals ryw ftr /kliuais riu u»r/ 'I heard some 
man', an immediate constituent analysis yields the following tree 
diagram i 





As I point out in section 2,2.r, gwr /guir/ 'man' undergoes soft muta- 
tion because of environment lie. Now if a noun had been the only 
element of NOUN PHRASE21 the noun would have undergone soft mutation 
because of environment Ilg. Taking this into consideration, we can say 
that environment Ilr is merely a subset of environment Ilg, which 
should read as follows « 'When the inflected form of the verb is used, 
any noun phrase which serves as direct object to the verb is in the 
environment Ilg. ' Accordingly, the initial segment of the entire 
phrase (if it is one of the affectable segments as shown in section 1.1 ) 
undergoes soft mutation. 

Such a revision contains a further generalization which serves to 
lend a greater economy to the description. The mutation environments 
do not affect words, as suchj rather, they affect entire constituents. 
The rules, then, do not apply to individual words that occur in the 
environment, but to the initial segment of the constituent. 

The notion that the rules apply to constituents and not to individual 
words is supported by the treatment of words in series. According to 
Morgan (1952*182), when a series of words occurs in a mutation environment, 


only the first word undergoes the mutation. In Morgan's example, there 
are three verb-nouns, canu /kani/ 'sing', dawns io /daunsio/ 'dance', 
and rhedeg /sedeg/ 'run', in a series in the sentence gallaf ganu, 
dawns io , rhedeg /gaiav gani daunsio sedeg/ 'I can sing, dance, run'. 
As the entire noun phrase, which includes the three verb-nouns, is in 
environment Ilg, the initial consonant of the entire phrase undergoes 
soft mutation. 

The relationship of environment I as a real environment can be seen 
in this phenomenon of embedded constituencies. In the sentence gwelals 
X Ay 11 /gwelais 9 din/ ' I saw the man ' , the fact that the definite article 
preceding a masculine noun constitutes an occurrence of environment I 
determines that no mutation can be effected. This lack of mutation 
cannot be reversed, no matter what mutating environment the word may 
enter at a higher level of constituency. 

In the generative framework, the constituency relationships shown 
in the application of mutation environments can be adequately handled 
through the transformational cycle (see Fillmore 1963). By cyclical 
application of rules, environments of embedded structures are established 
and rules applied, before the environments of the embedding structures 
are established. The mutation rules of Modern Welsh, then, undergo 
cyclical application, 

2.8 The Totality of the Environments . As stated in section 2,0, the 
mutation environments are both necessary and sufficient for the applica- 
tion of the mutation rules on affectable segments, as outlined in Chapter 
i. In the generative framework, these environments are established 
before the application of the rules. Thus, the environments have a 


decisive effect upon the segments which undergo mutation, but the seg- 
ments have no effect whatever upon the establishment of the environments. 

For example, in environment IV, the underlying initial segment 
|tl is realized as /G/, as in the word thad /9ad/ 'father (spirant 
mutation form)'. The realization of underlying Itl as /§/, moreover, 
is a clear indication on the surface that the affected word is in 
environment IV, If, on the other hand, we place the word mam /mam/ 
•mother' in the same environment, the underlying | ra } is realized /m/, 
giving us no outward indication (phonologically) that the word is in 
environment IV. Just because the result gives no outward phonological 
indication of environment IV, however, is no reason to suppose that 
environment IV is not present. Environment IV and all environments are 
syntagms which exist, in the generative model, prior to any phonological 
considerations . 

This point is particularly crucial to the description of the con- 
sonant shift and to the arguments presented in the next chapter. Each 
and every initial segment in the Welsh sentence is under one of the five 
environments at all times. These environments are the motivation for 
the realization of the underlying segments as surface (or intermediate) 
segments, and each environment may cause some segments to be changed and 
other segments to be realized in the radical form. Whether or not we 
see any change in the phonological specifications of a segment, the 
environment is present. 

The notion that every initial segment (at least) is in one environ- 
ment at any one time can be graphically illustrated as in Table 2.8, 
Environments II, III, IV, and V are specified in the grammar and occupy 
a restricted place in the universe. Environments II and IV overlap 


(see section 2.7), as do environments IV and V (see sections 2,4,b 
and 2.5). The rest of the universe is occupied by environment I. 
Environment I is, then, limitless, so that any grammatical string which 
does not meet the specifications of environments II, III, IV, or V 
must be included in environment I. There are no possible grammatical 
strings which exist outside the universe. 

Table 2.8 
Distribution of Mutation Environments 


il am using here phonological segments which axe reflected fairly 
closely in the standard orthography. Although w is written in the 
orthography, it is not a separate segmental 'sound' but is realized 
as rounding throughout the initial cluster of the word. 

2The verb-noun in Welsh has no English equivalent. It functions 
in many ways as a nonfinite form of the verb, but it enters into 
grammatical relationships as a noun. For example, in the periphrastic 
verbal construction, in conjunction with the predicator yn /an/ it 
is roughly equivalent to the English participle in the English peri- 
phrastic verbal construction, but when the object is a pronoun, the 
verb-noun is 'possessed' by the pronoun in the same manner in which 
a noun is possessed. 

3The exceptions are hyd /hid/, which (as cited) belongs to that 
class of prepositions governing the soft mutation but does not undergo 
a conjugation, and rhwng /run./ 'between', which does not belong to 
the class of prepositions governing the soft mutation but does undergo 
a conjugation, 

^There are, however, some exceptions. Chief among these are 
adjectives following the feminine noun nos /nos/ 'night'. These 
adjectives do not Undergo mutation. Hence, we find the phrase 
nos da /nos da/ 'good night* (rather than *nos dda */ nos & a /)» in 
which the adjective does not undergo soft mutation. For a more 
complete discussion of such phenomena, see section 3.0. 

5Note that in this case, the noun follows the adjective in 
environment He. This special type of relationship is treated further 
in section 2,7. 

^As is the case throughout the language, there are several dialect 
variants in use in different parts of Wales. For some examples of 
some of these, see R.O. Jones 1971. 

?At this point, I deviate from the order presented in James 
(1966:43). I do this in order to separate tra /tra/ from the nega- 
tives, as together they do not really constitute a single grammatical 
environment . 

8 Again, as in section 2.2.b and section 2,.3.b, the contracted 
form of the possessive pronoun is added, ont the vowel ending in a 
limited number of words. Compare the distribution of 'w /u/ as a 



third person singular feminine possessive pronoun contraction and as 
a third person plural possessive pronoun contraction with the dis- 
tribution of the masculine pronoun contraction in 2.2,b and the feminine 
pronoun contraction governing the spirant mutation in 2.3.h. 


3.0 Exceptions and Counter-Examples . The environments which I posit 
in Chapter 2 for the various mutation rules include the most common 
of the mutation environments. The few which remain, however, do not 
constitute any particular problem for the description and can simply 
be added to the list (compare, for example, Evans and Thomas 1968i^51-2j 
Morgan 1952). There are no additional environments which can cause any 
discrepancies in the relationship which holds between the phonological 
side and the grammatical side of Welsh mutation. 

Notwithstanding the fact that no particular problems would be 
generated by the addition of other environments, there are indeed some 
problems which exist within the environments themselves. Most of these 
problems can be treated as belonging to one of two classifications — excep- 
tions and counter-examples (in the manner they are used in Hjelmslev 

1970 130-1). 

When a word would 'normally* form part of a mutation environment, 
either as a conditioning element or the item to be affected, and the 
mutation rule is not applied because of some arbitrary and idiosynchratic 
nature of this word, then the word is an exception to the mutation. In 
order for this to be a true exception, the fact that the word does not 
undergo mutation cannot be predicted from any morphological or syntactic 
(or, indeed, semantic) class to which the word may belong. 

I mention an example of one such exception in section 2,2. n. 



Although nos /night/ 'night* is a feminine singular noun and da /da/ 
'good' is an adjective, when da /da/ follows nos /nos/, environment 
Iln does not apply. Thus we find nos da /nos da/ 'good night', but not 
*nos dda Vnos 6 a/, in spite of the fact that the latter is what one 
would expect. 

Neither nos /nos/ nor da /da/ belongs to any grammatical or semantic 
class which regularly violates the conditions of environment Iln. We 
can say, then, that nos /nos/ (at least) in this case constitutes an 
exception to the mutation rules. It is not enough merely to point this 
out, however. In a generative description, this exception must be 
described within the system. 

The manner in which the exception is accounted for within the 
generative model is neither phonological nor grammatical. Rather, the 
exception is noted in the lexicon j that is, it is marked for the 
particular word in the base component of the grammar (see Chomsky 
1965)t Now the sum of all grammatical features that determine the 
manner of lexical insertion and permissible transformations for each 
word In the lexicon is found in the complex symbol of that word. Thus, 
nos /nos/ would be designated as [+Noun, +Comaon, -KSount,...!. 
Assuming that it is this noun that forms the exception to environment 
Iln, we would add the feature [[-environment Iln] to the complex symbol 
of this word. 

It is important for our description that we appreciate the implica- 
tions of our including a blocking feature in the complex symbol. The 
lexical subcomponent is found, as noted above, in the base component 
of the grammar. As the transformational -generative model is directional, 
we thus Insure that the necessary information is noted in the grammar 


before the application of transformational rules in the syntactic 
component. This is necessary, for the transformational rules estab- 
lish the various environments noted in Chapter 2. Now that the blocking 
feature is specified, as soon as the transformations yield environment 
Iln, it is known that the mutation rules in the phonology (in this case, 
rule 1.1. b) will not be effected. 

At least in the approach to transformational grammar in which 
lexical items are inserted before transformational rules take place 
(see Chomsky 1972), the matter of exceptions to the mutation rules 
presents no problem. Exceptions are simply lexical items with particular 
blocking features in their complex symbols. A similar situation can 
be found in New High German, in which the word Nacht /naxt/ •night', 
which is marked [+femlnine], In certain grammatical conditions (the 
indefinite time adverbial) is marked [-feminine] in order to derive the 
phrase eines Nachtes /aines naxtes/ 'one (indefinite) night'. Indeed, 
the effect of blocking environment Iln in the case of the Welsh nos 
/nos/ would be the same as we find in the German example. 

Whereas an exception is idiosynchratic In nature, a counter-example 
represents a particular pattern. When a delimited class of items 
regularly fails to activate the mutation rules, then the class forms a 
counter-example to these rules. Now the basis of classification can be 
grammatical or semantic, but a basis that can be defined must exist. 

Perhaps the most common example of a counter-example in Welsh 
mutation is the proper noun. Although a proper noun may, in some 
Instances, undergo mutation (compare Morgan 1952 «3)» it usually does 
not, when the proper noun is the name of an individual. Thus, a sentence 
such as gweloddMair /gweloo mair/ 'he saw Mary' is grammatical in spite 


of environment Ilg, 

As in the case of the exception, the counter-exanple is narked in 
the lexicon. Rather than marking every personal name, however, we can 
nark the entire class by a lexical redundancy rule by which the features 
[-Common] and [-♦Personal] imply the feature [-Mutation] in the complex 
symbol. Although we thus accommodate an entire class rather than a 
single item, the implications inherent in the counter-example for the 
operation of the transformational -generative model are the same as those 
inherent in the exception. The only difference in our description 
between an exception and a counter-example is that one entails the 
marking of a particular complex symbol while the other entails the 
marking of a class of complex symbols. 

Leaving aside the currently uncertain position of semantics, the 
present description makes use of (l) a base component, which supplies 
the lexical items with their particular markings and which supplies 
the basic string j (2) a syntactic component, which arranges the basic 
string of marked lexical entries into the various environments j and (3) 
a phonological component, which interprets the lexical entries relative 
to the syntactic arrangement and applies the mutation rules wherever 
applicable. Regardless of any exceptions and counter-examples, then, 
the basic relationships between the phonological half and the grammatical 
half of mutation can be described through rules which are regular. 

The regularity of the mutation system from base component to 
phonological component hinges upon the maintenance of the relationships 
that obtain between the syntax and the phonology. This is to say that 
all other problems which may arise can be handled by greater or less 
specificity in the lexical entry or by transformations, so long as the 


problems do not affect the basic relationships between grammatical 
environment and phonological rule. 

As stated at the outset of this chapter, most of the problems 
(such as personal nouns) which seem to exist in the mutation system 
can be treated as exceptions or as counter-examples. There is one 
other situation, however, which cannot be so treated. This case, in 
fact, appears to affect the basic relationships between grammar and 
phonology. Because the word gan /gan/ 'with' exemplifies this, I call 
this situation the problem of gan /gan/. 

3.1 The Problem of gan . The word gan /gan/ can function either as a 
preposition, with the meaning 'with' or 'by', or as a conjunction 
introducing an adverbial clause, with the meaning •am' or 'while'. 
Now it is often the case that an adverbial clause will be introduced 
by the conjunction a /a/ 'and' or by its negative na /na/ 'and not'. 
As we see in section 2.4.f, this conjunction (and its negative) is 
sufficient motivation for environment IVf and the application of the 
spirant mutation rule, if the initial consonant of the following word 
is a voiceless aspirated stop (or aspirata). 

Thus, when the conjunction a /a/ precedes the word gan /gan/, 
we should expect the resulting construction to be *a gan */ a S 8 ^/ ' t 
because, although one necessary factor in the mutation is present (the 
conjunction a /a/), gan /gan/ does not have a voiceless aspirated stop 
in initial position. The fact is that Igl does not undergo spirant 
mutation and should be realized here as the radical /g/. However, 
*a gan */ a S^W is not the grammatical construction. The acceptable form 
in Standard Welsh is a chan /a xan/. 


The problem in the relationship between gan /fan/ and a chan 
/a xan/ is a fairly obvious one. An element of mutation environment 
IVf causes a segment which normally does not have a different spirant 
mutation form either to create a special spirant mutation form or to 
borrow the spirant mutation form of some other segment. Throughout 
Chapters 1 and 2, we see that it is a basic principle of operation that 
the mutation system consists of regular phonological rules motivated 
by distinct environmental factors. The exceptions and counter-examples 
of section 3.0, moreover, do nothing to weaken the relationship between 
grammatical environments and regular phonological rules — they merely 
demand a greater amount of lexical marking. Now this situation brings 
about a problem in the basic relationship between grammar and phonology, 
and even within the phonology itself. 

As we attempt to find a solution to this problem which will allow 
us to include the phenomenon in an explicit generative description of 
Welsh, we ought to realize that a simple lexical specification cannot 
provide us with the solution. In the cases of exceptions and counter- 
examples, all that is necessary is some mark which blocks the application 
of the phonological rule. In this case, on the other hand, we do not 
find an exception or a counter-example, but we find a deviation. It 
is first necessary for us to determine just what it is that is deviating 
and how it is deviating, before we can add any specifying mark to the 

There are two other prepositions that undergo the same deviation, 
and I would like to mention them here. Gyda /geda/ 'along with 1 and 
ger /ger/ 'at' are realized in environment IVf as a chyda /a xada/ and 
a cher /a xer/, respectively. It Is not, then, a simple case of one 


isolated word, but a case of several words which may fora a sub-class of 
the function words. The fact that function words, words of high incidence, 
are Involved, of course, heightens the necessity of accounting for the 
deviation in a generative description. 

3.2 The Minor Rule Based upon Iganl . At the center of the problem of 
gan /gan/, we do not find a grammatical deviation in and of itself. 
The graiuaatical considerations inherent in environment IVf have not 
changed—only the phonological rules effected by the environment have 
changed from their specifications in Chapter 1. Thus, any lexical 
feature added to the item Iganl will have to refer to a specific phono- 
logical rule to be invoked in these particular instances. The problem 
is to determine what rule to posit. 

As we observe the surface phenomena, we can see that a phonological 
/g/ alternates with a phonological /x/. As I demonstrate in section 
1.0, the first phonological segment of /gan/ is the radical segment, 
the segment which appears in environment I, and because it is the 
radical, it is by definition the underlying segment | g| from which both 
/g/ and /x/ are derived. Thus, quite simply, the phonological rule to 
be marked in the lexical entry is one which derives /x/ from Igl in 
environment IV, 

However, there appears to be a major complication. According to 
the mutation rules, only /g/ can be derived from |gt in environment IV 
(compare section 2.8). Thus, we have one environment for one underlying 
segment, but two phonological rules applying in that one environment. 
This creates a complication, for we must now find some way of telling 
when to use one rule and when to use the other. In order to overcome 


this complication in our phonology, we can make use of the notion of a 
'minor rule'. 

A minor rule is a rule with limited range of application which is 
applied in place of a (major) phonological rule when the lexical entry 
of the particular item specifies that the minor rule be used. Thus, 
the lexical entry for Iganl could be marked [-tMinor rule] so that the 
normal mutation rule deriving /gan/ would be superseded by the minor 
rule deriving /xan/. 

This minor rule can be posited as follows i 

a. i«i - h i iv [wj . |£g / gg iv 

From the outset, this minor rule appears suspect. As we can see 
from the mutation rules in Chapter 1, the tendency in mutation is for 
voiceless aspirated segments to become voiced unaspirated segments, not 
for voiced segments to become voiceless. In section 2.1, moreover, 
the hard mutation, which is also allegedly a case of voiced becoming 
voiceless, is discounted because of this very tendency. As voiced 
segments do not regularly become voiceless obstruents, the minor rule, 
as it is found in rule 3. 2. a, does not appear to be consistent with 
the system. 

Moreover, in this minor rule, we find two major feature changes- 
voiced to voiceless, noncontinuant to continuant. Although this is not 
without precedent (compare rules 1.1. d and l.l.e), most phonological 
rules treated thus far in the literature change only one feature specifica- 
tion at one time. 

This latter point may give rise to an intermediate stage hypothesis. 
By using an intermediate level in our derivation, we can first derive /y/ 


fromlgl, a process which would maintain the frication process of 
spirant mutation. As this /y/ is not in the inventory of Welsh sounds, 
it can be realized as the closest segment to it — /x/. This would result 
in the following minor rulei 

b. Igi - Nl - M I iv 

[+vcd"| „ Pvcdl _ pvcdl / r+obs] 
^aspj £cntl IjcntJ ' l+bckj IV 

As mentioned in section 1.1, however, the intermediate level makes 
use of a 'false step*, the positing of a segment which never appears 
on the surface. As we see in that section, the process by which a 
voiced unaspirated stop is realized as a voiced fricative is not at 
all indicative of spirant mutation, but it is indicative of soft muta- 
tion. Were we to allow the false step derivation of intermediate /y/ 
in this Instance, moreover, there would be no reason to disallow it in 
a more general form of mutation rule 1.1. b. Such a situation would 
have us derive null from /y/ in the one case and /x/ from /y/ in the 

I should like to leave environment IV momentarily in order to 
examine the behavior of gan /gan/ in environment II. Morgan (1952 »385) 
mentions the string £ gan /e gan/. Assuming that the minor rule 
necessary to prohibit the derivation of *yr an (by rule 1.1. c, as in 
y_ gan /a gan/) acts upon the underlying Iganl, then we mast posit a 
minor rule such as the following t 

C Igl - Id I n 

This minor rule looks suspect, for the soft mutation rules are 


the broadest of the nutation rules in scope (see also Griff en 197^b). 
It seems irregular that a minor rule would be needed in effect to block 
the application of a soft nutation rule. By using the intermediate 
stage of rule 3.2.b, however, we can invoke a change in soft mutation, 
as follows i 

d. Igi - HI - lei I ii 

Of course, such a minor rule as 3.2.d is clearly absurd. It takes 
the premise of the false step beyond reason, and it certainly destroys 
the intermediate /y/ hypothesis. However, if we observe rules 3.2. b 
and 3.2.d, an alternative intermediate stage hypothesis suggests itself. 
The alternative intermediate level contains the phonological segment /k/. 

The use of the intermediate /k/ should eliminate the unnatural 
nature of the intermediate /y/, for we know that /x/ is in fact derived 
from underlying Ik I and that /g/ is in fact derived from underlying 
|k| in the regular application of mutation rules 1.3 and 1.1. a. All 
that Is needed in the new minor rule is the following » 

e. Igl - M I 11,1V 

Rvcd] _> pvcd! / [+obs| 
L-aspJ baspj ' |-cnt| 


Rvcdl _> Pvcd] I f+obsl 
L-aspJ baspj ' -cnt 

L+bckJ II,: 

Such an hypothesis, however, is not credible because there is no 
rule relationship between the |g| and the /k/. Instead, what credibility 
there may be in minor rule 3.2.e lies in the fact that, when it is 
ordered before the mutation rules, we need only apply rule 1.3 in environ- 
ment IV and rule 1.1. a in environment II. Thus, we find the following 
systematic distribution. 


f. Id I ii 
Igl - M< 

M / iv 

In spite of the credibility lent to it by the subsequent nutation 
of the intermediate /k/, the minor rule 3.2.e runs counter to the 
tendencies of the mutation system by deriving a voiceless aspirated 
stop from a voiced unaspiratedstop. 

Moreover, as I show in the next chapter, if one of these words 
were to be found in environment III, the resulting phonological seg- 
ment would also be formed from the intermediate /k/ (compare the form 
ynghyd /eghid/ 'together', related to gyda /geda/). Of course, such a 
development would also require minor rule 3.2, e and produce the follow- 
ing systematic distribution » 

g. /g/ / II 

Igl - A/£7gh/ / in 
M I iv 

At first glance, the systematic distribution 3.2.g looks fairly 
regular. However, only three of the four environments which affect 
initial consonants are incorporated into this statement. In environment 
I, the minor rule does not apply, so that /g/ can be derived directly 
from |g/. Taking this into account, our distribution statement may be 
written as follows i 


h. /g/ /I 

^<C /g/ 7 n 

* /gh/ / in 
H I iv 

This distribution statement should render one fact obvious. By 
itself, the underlying | g I without the intermediate /k/ can account 
for only one phonological segment. The intermediate /k/, on the 
other hand, can account for three of the four phonological segments. 
A much simpler solution would derive the /g/ in environment I also 
from the /k/, a practice which I investigate in the next section. 

3.3 The Minor Rule Based upon lkanl . As we see in section 3.2, by 
using the seemingly obvious assumption that the initial consonant 
found in environment I corresponds to the underlying segment in all 
instances (an assumption valid throughout the mutation system to this 
point), we arrive at a solution to the problem of gan /gan/ which is 
so complex as to be unnatural. Implicit In the analysis of section 
3.2 is the notion that the grammatical considerations of mutation 
can bring about the change of an underlying segment so that the muta- 
tion rules can produce certain distinctions in the output of the 
phonological component. Such a notion represents a departure from the 
regular relationships between grammar and phonology which is so great 
as to render this analysis (section 3.2) at best suspect, unless there 
is further strong corroboration. 

As stated at the end of the last section, only one of the four 
ultimate phonological segments cannot be predicted from the intermediate 
/k/ using the already-existing rules in Chapter 1. This point raises 


the possibility of using the underlying segment |k) and then a minor 
rule deriving /g/ froa Ik I in environment I, Such a solution would 
increase the simplicity of our description considerably. 

Of course, one area of the description that can be simplified by 
using underlying Ikl is the systematic distribution. Statement 3.2.h 
represents a complex distribution requiring two rules for the derivation 
of each of three phonological segments. A minor rule such as 3.3 » 
below, however, would have two rules deriving only one of the four 
segments, if any. 

Of far greater importance to our description is the impact of such 
a minor rule on the mutation system itself. The mutation rules are 
morphophonological in the sense that they are motivated by grammatical 
considerations, irrespective of phonological considerations. A minor 
rule of the type found in 3.2. e would have mutation rules operating 
on an intermediate segment—not on the underlying segment, as we find 
elsewhere throughout the system. A minor rule operating on underlying 
Ikl, on the other hand, would maintain the notion that the mutation rules 
operate directly on the underlying segment. By thus allowing a greater 
degree of conformity with the mutation system, this alternative solution 
lends considerable simplicity to the system. 

For these reasons, then, as well as for others considered later 
in this section, we can posit the following minor rule: 

Ikl - /g/ / I 

f-vcdl _ r+vcdl / f+obsl 
L+aspJ L-aspJ ' +bck 


Although minor rule 3.3 is superior to minor rule 3.2.e, I include 
the complicated arguments in section 3.2 in order to demonstrate two 


areas of concern that might weaken minor rule 3.3. First of all, minor 
rule 3.3 operates in environment I, the unspecified environment. It 
is otherwise axiomatic that where environment I exists, no mutation is 
effected. In section 3.2, however, we see the complexity of description 
to which this supposed axiom would lead us. Furthermore, the minor rule 
is not, strictly speaking, a mutation rule. Thus, we can still main- 
tain that mutation is not effected in environment I. 

The concern over the nature of environment I leads directly to 
the second area of concern. This second area has to do with the notion 
of 'absolute neutralization 1 (see Kiparsky 1973). By using an underlying 
|g|, we maintain the 'alternation condition', avoiding absolute neutraliza- 
tion. This is to say that the underlying Igl is reflected in one of the 
phonological segments on the surface, thereby maintaining some overt 
justification for positing the underlying segment as Igl. An underlying 
Ik |, on the other hand, is not realized on the surface as a /k/, thereby 
failing to maintain this overt justification. 

Of course, it is fairly easy for the linguist to posit an abstract 
underlying segment which does not occur on the surface in the descrip- 
tion of an 'exotic' language (for example, Hyman 1970). Yet, any 
linguist who posits such an abstract segment in a well-known European 
language had best prepare a good defense. Welsh with its earlier forms 
boasts fourteen centuries of literary tradition, so this abstractness 
creates a particular area of weakness for the minor rule, if only by 
the fact that attacks against it will be strenuous. 

I should point out that minor rule 3.2. e also makes use of an 
abstract segment in the form of the intermediate stage. Minor rule 
3.3, then, is not really any more abstract than 3,2, e— it merely makes 


the neutralization into a more explicit statement. If we wish to 
avoid abstractness, we will have to derive the phonological segments 
/«/» /§ h /» and /*/ directly from |g|. 

As stated in section 1,2, voiceless aspirated nasals are only found 
as a result of nasal mutation. Moreover, the voiceless aspirated 
nasals can only be derived from their voiceless aspirated stop cognates. 
Indeed, this is the origin of the aspiration in the nasals. An occur- 
rence of /nh/ can therefore only be derived from underlyinglkl, not 

from underlying Jg|, 

Furthermore, as we direct our attention to the string a chan 
/a xan/, we find a voiceless velar fricative. This segment is not 
followed by an occurrence of /w/. Now it is extremely rare (if, indeed, 
it is even possible in literary Standard Welsh) that we find an initial 
/x/ which is not followed by /w/ except where that /x/ is the spirant 
mutation form of /k/. Thus, the speaker would assume, given the 
string a chan /a xan/ f that the /x/ is derived from |kl, not from |g|. 

There is also direct historical evidence to support minor rule 
3.3 and the underlying lk|. Although I do address this evidence in 
section 4,1, I do not believe it proper to introduce historical evidence 
at this point, as we are involved in these first three chapters with a 
synchronic description of the Welsh mutation system, and such a syn- 
chronic description should be able to stand on its own merits. On the 
other hand, there is some historical evidence which may be relevant to 
the synchronic description, for it pertains to a tendency which may very 
well exist in Modern Welsh, This tendency is lenition (see section 1,1 ). 
According to Jackson ( 1953* 23), the process of lenition is found in the 
fifth century A.D. in both Brythonic (>Welsh) and Goidelic (>Irish) 


due to some phonological 'nuance' inherited by both languages from 
Common Celtic. Indeed, both Welsh and Irish still maintain remarkably 
similar lenition. If this tendency, which was strong enough to affect 
two languages after they had separated from a common source, still 
exists in Modern Welsh, then a minor rule deriving /g/ from |k| is 
certainly not unexpected. I return to this tendency throughout the 
following chapters. 

In spite of the problems with absolute neutralization, the evidence 
seems to support minor rule 3.3. Clearly, the alternatives create more 
problems than they solve. However, there is still need for some strong 
corroboration — some evidence which supports the notion that rule 3»3 
adequately describes the deviation in the mutation system, 

3.4 Evidence for the Minor Rule . The analysis of the deviation through 
minor rule 3.3 in the previous section certainly leads to a simple, 
concise description of the phenomena involved. This description affords 
maximum generality and conformity with the mutation system. Finally, 
the description by way of a minor rule is explicit and adheres to the 
generative methods of description. 

In spite of the fact that this analysis provides a simple, general, 
generative solution to the deviation, the phenomena which are involved 
in minor rule 3.3 still constitute a deviation. A deviation must have 
corroboration — some further evidence from some other point in the 
linguistic system which supports the proposed solution to the deviation. 
By corroboration, I do not mean simple recitations of data which mirror 
exactly the deviation examined above. Instead, we need some evidence 
which more closely ties the phonological voiced unaspirated stop in 
environment I to the underlying voiceless aspirated stop, further 


justifying our analysis. 

In order to find corroboration for our minor rule, we need go no 
further than the class of prepositions itself. There are two pairs of 
prepositions crucial to this analysis. They are tros t dros /tros,dros/ 
'over' and trwy > drwy /truii,druii/ 'through'. 

In environment I, both members of each pair may be used in free 
variation. For example, the phrase tros y_ cae /tros a ka»i/ 'over the 
field' may be found in the same text as the phrase dros y_ cae /dros 
9 kaji/ 'over the field'. Moreover, the phrase trwy 'r drws /trutir 
drus/ 'through the door' may exist along side of the phrase drwy 'r 
drws /druiir drus/ 'through the door'. At this point, there is not any 
motivation for deriving the second member of each pair from the first, 
as a soft mutation environment is lacking. And there is certainly no 
motivation for deriving the first from the second, as this would contra- 
dict the 'normal' tendency found in the mutation system, 3 

It is in environment IV that we find the needed corroborating 
evidence. If the prepositional phrases mentioned above are preceded 
by the conjunction a /a/ 'and', then the preposition enters environ- 
ment IVf . We might expect, then, to derive the phrases a thros y_ cae 
/a 9ros a kaii/ 'and over the field' and *a dros y_ cae Va dros » kaii/ 
from the first pair, and the phrases a thrwy 'r drws /a eruiir drus/ 
•and through the door' and *a drwy 'r drws */ a druiir drus/ from the 
second pair. The second member of each pair, however, is not found. 
When either member of the two pairs enters environment IV, the spirant 
mutation form from the first member is derived, regardless of which 
member of the pair is used in environment I. 

The fact that only the spirant mutation form of underlying |t| 


is permissible in environment IV indicates that the underlying form 
of both members of the first pair is Itrosl and of both members of the 
second pair isjtruii|. In order to derive the optional second member 
of each pair, the underlying form must undergo a minor rule of the form 
found in rule 3.3 in environment I. 

The alternation of /t/ and /d/ both derived from Itl in environ- 
ment I is evidence that a voiced unaspirated stop can indeed be derived 
from a voiceless aspirated stop in the unspecified environment. More- 
over, the existence of /t/ along side of /d/ satisfies the alternation 

Perhaps a more interesting pair of this type is found in tan 
/tan/ and dan /dan/. Like gan /gan/, this pair represents a preposition 
meaning 'under' and a conjunction introducing an adverbial clause Kith 
the meaning 'while (in the past)'. The use of tan /tan/ as a preposition 
is fairly rare except in certain compound prepositions, such as oddl tan 
/o6i tan/ 'from under'. Tan /tan/ is more often used in the adverbial 
construction, where (as noted above) the conjunction a /a/ is often 
found with its environment IVf . Of course, when a /a/ does precede 
either tan /tan/ or dan /dan/, the resulting string is a than /a Gan/, 
The close grammatical and semantic relationship (adverbially) between 
tan,dan /tan, dan/ and gan /gan/ adds further credence to the minor 
rule hypothesis. 

Thus, the pairs tros , dros /tros.dros/, trwy , drwy /truti,druii/, 
and tan , dan /tan, dan/ supply the evidence needed to show that gan 
/gan/, gyda /geda/, and ger /ger/ all have an Initial underlying Ikl 
and undergo a now-more-general minor rule In environment I. 

This analysis is, in fact, suggested by Morgan (1952:^52). who 


claims that in these words the t /t/ and c /k/ are resurrected ( cael 
eu hadfer) after the conjunction a /a/ in order for the spirant mutation 
to be effected. Now Morgan is speaking in historical terms, but clearly 
the application of this 'resurrect ion' can be seen in the synchronic 

The minor rule now must be made more general in order to include 
the dentals as well as the velars. Moreover, the minor rule also 
affects labials, as shown in the next chapter. Thus the rule in its 
most general form can be written as follows t 

pvcd] _ f+vcdl / r+obs"| 
t+aspj LaspJ ' L~ cn y * 

This minor rule 3.4 can now be incorporated into the grammar of 
Welsh, The rule itself is in the phonological component, but the lexical 
items which undergo the rule are marked in the lexicon with the feature 
[*Minor rule 3, k]. Thus, the deviation can be accounted for in a 
generative description, 

3.5 The Synchronic Description . These first three chapters present 
a synchronic description of the Welsh consonant mutation system. 
Chapter 1 notes the regular phonological rules inherent in the system j 
Chapter 2 mentions the grammatical categories and constructions which 
motivate the system; and Chapter 3 deals with the exceptions, counter- 
examples, and deviations of the system and how these can be accounted 
for in a generative description. 

This description is explicit, in that the rules are posited and 
the methods for incorporating the rules are treated. On the other hand, 


this description is not intended to be exhaustive. However, additional 
information needed to describe the mutation system will not change the 
methods used here. Any additional environments need only be added to 
the existing classifications. Moreover, exceptions and counter-examples 
can be marked in the lexicon just as they are in section 3.0. Even if 
any other deviations may be found (which is doubtful) they can be 
handled in the same way as is the problem of gan /gan/ in this chapter. 
Phonological additions can also be incorporated if necessary. For 
example, as shown in Griff en 197^b, the inclusion of the affricates in 
the mutation system can be accomplished with the existing phonological 

As this has been a synchronic description, it has not been my 
concern how these rules and these environments have come into being in 
the Welsh language. I have only been concerned with how this system 
can be described at this point in time. With the deviation, however, 
we find a set of phenomena which demonstrates some degree of instability — 
two rules operating on the same form in the same environment. This 
instability of the deviation appears to be an indication of linguistic 

As the deviation supplies us with the element of historical lin- 
quistics, I leave the purely synchronic nature of the system and consider 
the historical aspects in the next chapter. It is this deviation which 
leads most directly into the matter of the new Welsh consonant shift. 


1-This form is only found in written Welsh based upon earlier 
texts. It becomes clear by the end of this chapter that the article 
y /a/ in this case brings about environment II — as opposed to other 
cases in which it does not—for example, the preposition y_ tu mewn i^ 
/a ti maun i/ 'outside of. 

2ln fact, it can be argued that we should add a separate phono- 
logical segment /xw/ with the feature [+round] to Table l.O.c, to 
handle con-mutation forms of the voiceless velar fricative in 
initial position. 

3To be sure, there is the matter of hard mutation, which I discuss 
further in section 5.1. 



k.O The Generalization of the Minor Rule . Up to this point, we have been 
concerned with a synchronic analysis. In such an analysis, it does not 
matter where the deviation found in the problem of gan /gan/ (in the 
previous chapter) originated. Nor does it matter how this deviation has 
come into existence, nor where it may be leading. It is enough in a 
synchronic analysis to show that the minor rule 3,k offers the best 
resolution of the deviation. 

It is pointed out in section 3 A that Morgan (1952 1 452) attributes 
the deviation of such words as gan /gan/ and dan /dan/ to the 'res- 
surection' ( adfer ) of c /k/ and t /t/, respectively, as radicals, from 
which we may posit the underlying segments. Apart from the fact that 
this reference to the single most authoritative work on mutation converges 
upon the minor rule hypothesis, in our synchronic analysis the opinion of 
Morgan is of limited application. 

The deviation characterized by minor rule 3.^» however, is not of 
particular interest in the synchronic description of Welsh, except for 
its effect upon some generative phonological points of theory (see 
section 3.3). Primarily, the deviation is of historical interest. As 
stated in section 3.5, the very fact that we have a minor rule operating 
in Welsh is an indication that there may be some historical developments 
taking place. This connection between the minor rule as an entity of 
description and historical change is a central part of historical 



linguistics as it is described through the generative school. 

Sound change comes about as a result of the reinterpretation of 
the acoustic signal by the language learner (compare AnttHa 1972 1I98), 
This is to say that the learner creates a phonology which is different 
from that created by the linguistic source. The difference between the 
learner's phonology and the source's phonology can be described in the 
generative mode through the existing theoretical devices. 

In phonological theory, as found in Chomsky and Halle 1968, Schane 
1973, and so forth, we begin with a 'deep' representation and apply 
rules to this representation and to subsequent intermediate representa- 
tions in order to derive the 'surface' forms. Thus, the generative 
method must always start with an item. In historical linguistics, this 
item is the underlying 'deep' form of the older dialect. An historical 
change involves a deviation from the derivation used in the older dialect. 

The change, then, in historical forms can be described through 
newer rules applying to older forms. According to King 1969, there are 
four classifications of change, as seen through the generative framework. 
The first is restructuring — the underlying form of the older dialect 
changes, without affecting the rules which derive surface forms. The 
other three classifications maintain the older dialect's underlying forms 
but change the rules that derive the surface forms from them. We may 
find rule addition (a new rule added to the derivation), rule deletion 
(an old rule becoming obsolete), or rule reordering (a rule which had 
appeared before another rule coming after it, thus changing the output 

It should be fairly clear that in this framework a rule cannot 
be gradually added or gradually deleted or gradually reordered. The 


change must be abrupt. Yet, as we look at past sound changes, we can 
see that the language does not abruptly change from, for instance, 
Brythonic to Old Welsh. The change itself comes about gradually through- 
out the language community. The gradualness of the change throughout 
the language community is explained by the notion that although the 
restructuring or rule change occurs abruptly, the spread of this change 
through the community is gradual. This abrupt change-gradual spread 
hypothesis is a major tenet of generative theory (see King 1969:119). 

A language at one stage can be differentiated from the same language 
at another stage through a description using restructuring and rule change. 
But what of the period between the two stages? Indeed, the heterogeneity 
of the language community is an indication that the language is continually 
undergoing change (see Weinreich, Labov, and Herzog 1968), 

It is in this period of on-going change that the notion of the 
minor rule is invoked. In the case of rule addition, for example, we 
find during the gradual spread of the added rule that the older rule 
is still in effect in some dialects. As there is no restructuring, we 
find two rules (assuming that derivation from one stage to the next 
involves a rule) both operating upon the same underlying segment in the 
language community. In order to account for this situation, we say that 
the newer rule is a minor rule which is in competition within the 
language community. This minor rule then spreads, by becoming the more 
prestigeous form, until it displaces the dder rule. (Compare also H. 
Anderson 1973.) 

Thus, within the generative approach to language change, the minor 
rule may be syratomatic of on-going linguistic change. On the other hand, 
it may very well be that a minor rule enters the language (by change, to 


be sure) but remains a simple exception to rules of the grammar. As 
an exception, such a rule would be entered for a limited number of words 
in the lexicon. As we see in section 3.0, for example, the fact that 
the phrase nos da /nos da/ 'good night' does not bring about the muta- 
tion rule 1,1. b because of environment Iln is not an indication of a 
change in the language by which environment Iln may be eliminated. It 
is simply an exception of no historical consequence. 

In order for a minor rule to be indicative of linguistic change, 
it must satisfy a crucial requisite of historical development—it 
must spread. So long as a minor rule is spreading, no matter how 
gradual that spreading may be, a linguistic change is occurring by 
definition. 1 

The spreading of a minor rule is known as 'rule generalization'. 
There are two ways in which generalization can take place in generative 
phonology. On the one hand, generalization is said to occur when a 
rule applies to more environments. This is central to linguistic 
change. On the other hand, a rule may generalize so as to affect an 
entire natural class (see section 1,1). This is also an important 
aspect of synchronic description as well as historical change. As we 
see in this chapter, both types of generalization affect the minor rule 
3.4, indicating that this is, indeed, a case of linguistic change. 

4.1 The Historical Development of the Minor Rule . In order to show 
that the minor rule represents a current process of linguistic change, 
it is necessary to demonstrate that the rule is indeed generalizing to 
more and more words. That is to say that it must be shown that the 
minor rule is a growing phenomenon, not simply a relic surviving from 


a now-unproductive historical accident. 

As I mention aove, Morgan (1952 «^52) attributes the deviation of 
gan /gan/, for example, to the 'resurrection' of the radical c /is./ , 
Likewise, the deviations of the other function words with initial 
velars can also be traced to this 'resurrection'. As we turn our 
attention to the historical development of the deviation, it becomes 
proper and necessary to consider where this 'resurrected* c /is./ comes 

In Old Welsh, gan /gan/ is found spelled as can or cant (see Morris 
Jones 1913i405f Jackson 1953^96} Evans 1960tl24). Thus, in the Old 
Welsh period (ninth through eleventh century) the radical c /k/ was 
used exclusively in environment I, There is therefore no reason to 
assume that the minor rule was in existence at this point in history. 

In Middle Welsh (twelfth through fourteenth century) , on the other 
hand, both can (or kan) and gan can be found in environment I (see 
Morris Jones 1913 i 405? Evans 1960j12^), The introduction, then, of the 
minor rule for the velars occurred during the Middle Ages. Even in 
the Early Modern Welsh poetry, moreover, the radical c /k/ is sometimes 
found (Morris Jones 1913 1^). 

This history of gan /gan/ (and the others) indicates that the 
minor rule came into existence in the Middle Ages and competed with 
the major derivation into the Early Modern Welsh period (fifteenth 
through sixteenth century), at least in the more conservative verse. 
In Modern Welsh, however, the minor rule has totally eclipsed the older 
forms such that *can is not even listed in Evans and Thomas 19^8, in 
spite of the fact that many obsolete forms are included with asterisks 
in that work. 


The extent to which the minor rule became dominant in the Middle 
Welsh period is open to some doubt. The uncertainty arises from the 
fact that the consonant mutations were not regularly designated in 
Middle Welsh orthography (Evans I960 t 9). Thus, a scribe interpreting 
the minor rule for the soft mutation could very well write can for 

An interesting aspect of the growth of the minor rule throughout 
these periods lies in the two different functions of gan /gan/. As 
noted in section 3.1, this word functions as a preposition and as a 
conjunction introducing an adverbial clause. There is reason to believe 
that the minor rule dominated the word in its prepositional function 
before it dominated it in its adverbial function. According to Morris 
Jones (I913i443), it does not appear to undergo the minor rule in its 
adverbial function in Middle Welsh. Furthermore, the Modern Welsh 
conjunction canys /kanis/ 'since' is still found, and it is a contraction 
of the Middle Welsh can 'since' and yjs 'it is' (see also Morgan 1952 i54l). 

Thus, we can set the following time-table for the domination of 
gan /gan/ by the minor rule 3.4, The minor rule becomes an optional 
variant of the standard derivation in the early Middle Welsh period. 
During this period, it gains prestige over the prepositional form, but 
it lags in the adverbial form. By the Early Modern Welsh period, the 
traditional form of the preposition only survives in the conservative 
verse poetry, and it begins to disappear in the adverbial function. 
By the Modern Welsh period, the minor rule is obligatory. 

So far, I show that an historical progression of the minor rule 
is traceable through the Early Modern Welsh period. But in itself, 
this does not tell us much about the language today. The question 


remains as to whether or not the minor rule is still a viable factor 
in Modern Welsh. 

In order to relate the generalization of the minor rule to contem- 
porary Welsh, it is only necessary to cite the pairs of prepositions 
tan , dan /tan, dan/, tros , dros /tros,dros/, and trwy , drwy /truii.druii/, 
mentioned in section 3.4, Yet, the fact that minor rule 3.4 is optionally 
Invoked in these cases does not prove that the rule is still spreading — it 
is possible that the entire process has been frozen since the Early 
Modern Welsh period. Indeed, these alternations were noted in 1567 
by the Welsh grammarian Gruffydd Robert (1927 »3» 39-40 )» who wrote that 
the words tros , trwodd , traw, cann , cidag were more usually spoken as 
dros , drwodd , draw , gann , gldag (his orthography). Thus, we need some 
more evidence to show that the generalization (the spreading of the 
rule to new environments) is still, in fact, active. 

As mentioned in section 3.4, an interesting parallel to gan /gan/ 
is found in the pair tan , dan /tan, dan/. It is interesting because of 
the fact that both function as prepositions and as conjunctions intro- 
ducing adverbial clauses. In their adverbial functions, moreover, 
they are semantically quite similar. 

In Old Welsh, tan, dan /tan, dan/ is in a radically different phono- 
logical shape. It is written as guotan or gutan (see Morris Jones 
1913*3991 Jackson 1953 «389? Evans 1960il35). Clearly the t is found 
in this period, although it is not in the environment of the minor 
rule. In Middle Welsh, the alternation tan~dan is found extensively 
(Morris Jones 1913*399* Evans 1960:135) » as it is in Early Modern Welsh 
(as shown by the reference of Gruffydd Robert, above). 

It is in the contemporary Modern Welsh that we find a significant 


development. This development can be demonstrated by referring to two 
recent textbooks — James 1966 and Bowen and Rhys Jones i960. The advan- 
tage of citing a textbook is that a textbook is usually more internally 
consistent in its use of the language than is a body of data (such as 
Fynes-Clinton 1913). Thus, although it may not be an accurate represen- 
tation of what is actually said, it does represent what one scholar 
considers to be the more usual forms, at least in that scholar's usage. 
We may consider the textbook, then, to be a comment on what is considered 
the 'norm'. It is important for us, whenever we consider what is written 
in a textbook, to consider first just what kind of language that text- 
book represents. In Welsh especially, we must determine a relative 
degree of conservatism — whether the lnaguage used in the textbook 

represents a more conservative (closer to the Standard) or a less con- 

servative (closer to the colloquial usage) approach to Welsh. 

According to both James I966 and Bowen and Rhys Jones i960, dan 
/dan/ is the exclusive form of the preposition. The more conservative 
James 1966 lists tan /tan/ as the only form of the adverbial clause 
conjunction. The less conservative Bowen and Rhys Jones i960, on the 
other hand, lists dan /dan/ as the only form of the adverbial clause 
conjunction. Although the latter accepts only dan /dan/, both members 
of the pairs tros , dros /tros,dros/ and trwy , drwy /truii,druii/ are 
cited as acceptable. 

When we compare the discrepancy between the more conservative and 
the less conservative approaches to Modern Welsh with the time-table 
for the domination of gan /gan/ by the minor rule, we can discern one 
important fact. Dan /dan/ appears to be at that point of development 
today at which gan /gan/ was found in the Early Modern Welsh period— some 


five centuries ago. Thus, the process of generalization of the minor 
rule 3.4 is not only still active, but it is also following the same 
orderly progression as it did in Early Modern Welsh. 

There are no prepositions which begin with a labial stop, but 
there are some other function words. In 1913» Morris Jones noted that 
here too there were voiced-voiceless pairs, as in pie /pie/ and ble 
/ble/ 'where (interrogative)' and plau /pitai/ and blau /bitai/ 'whose 
is (are) 1 . According to Morris Jones (1913*359) » in the spoken language 
both /p/ and /b/ are heard, and the /p/ is particularly heard in North 
Wales, which is considered the more conservative dialect region. 

The pair pie, ble /pie, ble/, for example, could represent an 
alternation between two separate forms (as in Hoenigswald i960) or it 
could represent the presence of the minor rule. In the former case, 
mutations of hie /ble/ would be formed from an underlying lb I, and this 
alternative form would be a case of restructuring (or relexicalization) . 
In the latter case, ble /ble/ would behave like gan /gan/ or dan /dan/, 
forming its mutation forms from the underlying voiceless aspirated stop. 

Bowen and Rhys Jones (1960«29) lists ble /ble/ as the contracted 
form of pa le /pa le/ 'what place', but it does not list pie /pie/ 
as an alternative form. In environment Hid, however, it gives the 
contraction of vn /an/ and ble /ble/ as ymhle /emhle/ 'in what place', 
a clear indication that it is the minor rule 3.4 which is causing a 
deviation of the same type described above. 

Thus, the minor rule is an on-going process which is generalizing 
throughout the function words of Modern Welsh. Moreover, it not only 
affects velars and dentals, as Gruff ydd Robert noted in 1567 1 but it 
is now affecting labials. 


4.2 The Generalization to the Natural Glass . As pointed out in section 
k.O, generalization can occur with respect to the affectable environments 
of a rule. This is occurring as the minor rule affects the labials, for 
now the rule not only applies to prepositions and adverbial clause 
conjunctions, but it also applies to other function words as well, such 
as question words and the possessive relative pronoun-verb. Moreover, 
generalization can also occur with respect to a natural class of seg- 
ments. Again, this is occurring, as the minor rule now affects labials. 

Now the generalization of the minor rule such that it affects 
an entire natural class is an important development, especially from 
the standpoint of generative theory. Such a development involves the 
•simplicity metric'. According to King 1969» languages tend to change 
in such a way as to become more simple. In order to understand the role 
of simplification in the generative approach to historical linguistics, 
it is necessary first to consider the general framework employed by the 
generative phonologist. 

Generative phonological rules are formulated in features. Each 
feature tends to limit the scope of the particular rule. When we remove 
a feature, then, the rule applies to more segments, thus becoming more 
general. It is important to note that the process by which the rule 
becomes more general is precisely the same as the process by which the 
rule becomes simpler — the removal of an item. 

According to King 1969, languages tend to become simpler. Thus, 
the rules within the languages tend to become more general—applying 
to more items. This notion is closely allied to the traditional 
approach of analogy. 

As a Welsh dialect accepts the affricates /c/ and /3/ from English, 


for example, the mutation rules must add the feature [-strident] to 
block the mutation rules from applying to the affricates. In several 
dialects, soft mutation has generalized so as to affect the new affricates, 
and this is reflected in the removal of the feature [-strident] from the 
rules. In the Dyffryn Nantlle dialect (see R.O. Jones 1971 ), moreover, 
the same has happened with respect to the nasal mutation rules, resulting 
in the feature [-strident] being removed from these rules. As pointed 
out in Griffen 1974b, the removal of the feature simplifies the rule 
and reflects greater generalization within the grammar. Moreover, as 
the rules can now be written with fewer features and the plosives and 
affricates are phonetically similar, the generalization and simplifica- 
tion reflect inclusion of affricates into a natural class of stops, at 
least in the affected dialects. 

The first segments that appear to be affected by the minor rule 
are the velars, A rule affecting only the velar stops can be posited 
as follows t 


pvcd] _ Bvedl / [+obs [ 
KaspJ Iraspl ' +bck 

L-cntJ I 

This is identical to rule 3.3.^ 

As the dentals are affected, the rule can be posited as follows i 


pvcd] _ phrcdl 
|+aspj [-asp] 



Rather than reducing the number of features, this rule 4.2.b has increased 
the number of features over 4. 2, a. This does not mean that the rule has 
become less general — indeed it has become more general, for it now 
affects two positions of articulation rather than one. 

As the labials are affected, however, we find an unmistakable 
simplification, as follows: 


f-vcdl _ Pvcdl / [+obs] 
[+aspj j^aspj ' L- cn ^J 

Now the minor rule has reached its simplest form. Furthermore, as we 
consider the requisites of a natural class (see section 1,1) , we can 
see that the rule now affects an entire natural class. 

As it now stands, this rule (identical to rule 3.4) rewrites all 
voiceless aspirated stops as voiced unaspirated stops. As such, then, 
it changes one natural class into another. When we couple this fact 
together with the fact that the minor rule is an on-going phenomenon, 
we might ask what would be the logical extension of these developments, 
I address this matter in the next chapter. 


*I return to the point of this definition in Chapter 6. 

2 As can he seen in its consistent use of the two-part negative 
(nid. ..dim ( ddim ) , as opposed to nld. ..), Bowen and Rhys Jones i960 
is less conservative than James 1956" (see Pilch 1971). 

3The specification [+obstruent] can replace the features to the 
left of the arrow. I have not done this, because I want to maintain 
the parallel structure between these rules and the mutation rule 1.1. a. 



5.0 Sound Change and Consonant Shifts . As stated in section k.0, 
sound change comes about as a result of the reinterpretation of the 
acoustic signal. Of course, any major deviation in the interpretation 
of sounds in the language learner from the accepted norm of the learner's 
source (whether as a result of the failure of the source to articulate 
in a manner that would make the sound system of the source clear to the 
learner or as a result of conscious alteration of the data, as is 
typical in abduct ive situations) leads to the learner constructing a 
phonological system which is not identical with that of the source (if, 
indeed, two systems can be identical). Insofar as particular sounds in 
the source phonology do not correspond to those in the learner's phonology, 
we can say that there is a change in these sounds in the language. Thus, 
if the source uses the segment /t/ in a particular environment and the 
learner interprets this as the segment /d/, then by convention we say 
the the /t/ is being changed to /d/ in that particular environment. 

Now it is crucial that we recognize the difference between sound 
change and the conventionalized description of sound change. The actual 
event of a sound change involves the transmission of the language from 
one generation of speakers to the next. In reality, there is no change 
involved here at all — just the establishment of correspondences between 
one generation and the next. Such correspondences exist within the 
language so that the speakers of one dialect may understand the speakers 



of other dialects (compare Anttila 1972 t48j H. Anderson 1973). 

In the generative framework, as pointed out in section k,0, one 
usually treats the older form as the underlying form (or at least as 
an intermediate form) and derives the newer form from it. In deriving 
the newer form, the linguist notes that there has been a change in the 
underlying form or in the rules deriving the surface form between the 
previous description and the newer one. The nature of the change gives 
the linguist a classification of the change consistent with the model. 
This method may describe the actual event, but it is not to be taken 
as being the event itself. 

A sound change can be limited to a single segment. For example, 
Old Welsh /f/ (mentioned in section 1.0) corresponds to Modern Welsh /v/. 
Thus, we say in our description that /£/ has become /v/. 

On the other hand, a sound change can include all members of a 
natural class. In section k,2 t for example, we see that the minor rule 
involved in the deviation outlined in Chapter 3 affects not isolated 
segments, but all members of the class of voiceless aspirated stops 
(aspiratae) in a particular environment. In fact, this sound change 
involves rewriting all members of one natural class (the aspiratae) as 
cognate members of another natural class (the mediae) in a particular 

Such a sound change, in which one natural class is rewritten as 
another natural class, is known as a consonant shift. Of course, the 
notion of a shift has meaning only in relation to the description of 
language. It describes the fact that in one generation where we find 
/p/» A/» and A/t in some subsequent generation we find corresponding 
/b/, /d/, and /g/. The phonological change Involved in this consonant 


shift In which voiceless aspirated stops become voiced unaspirated 
stops can be described in the generative model by using the minor rule 
3.4, which would in the process of the shift become a major rule. 

The last point, that a consonant shift in which voiceless aspirated 
stops become voiced unaspirated stops can be described by minor rule 
3,4, should not be taken lightly, for its implications are quite 
far-reaching. By such a statement I mean that in any language with a 
phonological system in which there are voiceless aspirated stops cognate 
with voiced unaspirated stops, if the former natural class should 
change to the latter, then that exact rule which I designate as minor 
rule 3.4 applies in the description. This quality of minor rule 3.4 
to apply in similar situations in all languages is a result of the 
generative interpretation of features as universal entities. Thus, the 
feature [-voiced"), for example, is the same for Welsh as it is for 
English or Chinese. Moreover, the feature [-voiced] is the same in 
initial position as it is in medial or final position— a point central 
to the argument of this chapter. 

As we view the relationships between consonant shifts and phono- 
logical environments, we can consider two environment types conditioning 
the shift (provided we have any shift at all). We can have a shift 
with a restricted environment. That is to say that the shift from, 
for example, voiceless aspirated stops to voiced unaspirated stops 
takes place in some environments, but not all. For example, in Danish 
these two classes only contrast in word-initial position (Fischer- 
J^rgensen 1954), having been neutralized in other environments as a 
result of earlier shifts. 

Generally, however, when we use the term consonant shift, we imply 


a sound change Involving a natural class of consonants taking place 
with an unrestricted environment. Although the shift with an unrestricted 
environment may be affected by exceptions (individual lexical items), 
such a shift represents the notion of •regular' sound change. 

I use the minor rule 3.4 as an example of a sound change involving 
a natural class, and possibly as the basis of a consonant shift. Phono- 
logically, however, minor rule 3.4 is identical to soft mutation rule 
1.1. a. This raises a question as to why rule 1.1. a is not to be considered 
evidence of linguistic change. The reason why it is not to be considered 
as such lies in its grammatically restricted environment. The application 
of the soft mutation rule 1.1. a signals grammatical information by the 
fact that the voiceless aspirated stops contrast with the voiced unas- 
pirated stops. The presence of contrast in the system, then, and the 
grammatical function of this contrast serves to maintain the functional 
distinction between the two natural classes. In the case of a consonant 
shift in which one class merges with the other, as we see as the result 
of minor rule 3.4, the maintenance of distinctive function within 
these classes is necessarily suspended. 

Thus, the effect of an unrestricted consonant shift in the phonological 
system of a particular language is the suspension of contrast between 
the two natural classes of consonants involved. This does not mean that 
all voiceless aspirated stops, for example, disappear from the actual 
acoustic data. It means, rather, that if a speaker utters a voiceless 
aspirated stop after a shift from voiceless aspirated to voiced unaspirated 
stops has taken place, then such a segment does not serve a distinctive 
function in the phonology (compare the Danish example cited above). 
In generative terms, a consonant shift is phonological in that it may 


be described as altering the systematic (and even taxonomic) phonemic 
structure of the phonological component, although the systematic phonetics 
may produce an obsolete segment in free variation with the newer form. 

The diachronic approach to the spreading of the minor rule 3.4 
in Modern Welsh, which I treat in the previous chapter, has considerable 
implications when combined with the notions of the consonant shift. 
First of all, as we see in section 4.2, the minor rule affects an entire 
natural class— changing (in the description) voiceless aspirated stops 
to their voiced unaspirated cognates. As we see in this section, such 
a rule changing one natural class to another in the unmarked environ- 
ment (environment I) serves to suspend the contrastive function of these 
two classes and may be an element of a consonant shift. 

Furthermore, the suspension of contrastive function between the 
two classes is not an isolated event. As we see in section 4.1, the 
environment of the minor rule is gradually spreading (compare King 
1969«119) among function words (and, as pointed out in section 5.3» 
this spreading may apply to some content words as well). Such a develop- 
ment may, then, represent a shift in the restricted environment of word- 
initial position. 

In most cases, a consonant shift refers to a change in consonant 
classes in an unrestricted environment. In the generative framework, 
for the change inherent in minor rule 3.4 to be considered a consonant 
shift in a restricted environment, the rule would have to spread to all 
items in word-initial position. In order for minor rule 3.4 to represent 
a consonant shift in an unrestricted environment, the rule would not only 
have to spread to all items in word- initial position, but it would also 
have to spread to all environments of all items. 


Of course, there has not been a complete shift parallel to minor 
rule 3,4 in the Welsh language. On the other hand, there may be a 
shift in progress. In the generative framework, in order to prove that 
such a shift is indeed in progress, two pieces of evidence are necessary 
and sufficient. Firstly, there must be a single rule changing an entire 
natural class (in the inventory of underlying segments). Secondly, 
the rule must be generalizing to all environments. 

These two requisites of our proof necessarily depend to a great 
extend on the universal quality of features. This quality enables us 
to use the same minor rule (insofar as the actual phonological change 
is concerned) in all of these environments and makes a shift evident 
by the very use of the same rule in the various environments. 

One matter must be stressed. As we cannot analyze the workings 
of the speaker's mind, our proof is necessarily limited to his performance, 
and more precisely to our description of his performance. The danger 
that lies herein should be evident. The proof is in the manner of 
describing the performance, but the description is not the event. I 
return to this matter in the next chapter. 

5.1 The Position of Neutralization . The minor rule does affect a limited 
environment in word-initial position (Chapters 3 and 4), and I return to 
that position in section 5.3. I should now like to examine the realization 
of stops in word -final position. 

Because I treat the notion of •hard mutation' in section 2.1, 
this phenomenon is perhaps the best avenue through which to approach 
the role of word-final stops. In this phenomenon, a word-final voiced 
unaspirated stop alternates with a medial voiceless aspirated stop. 


Thus, teg_ /teg/ 'fair' has the equative form teced /teked/, for example. 

As I point out in section 2,1, the more general solution to this 
problem, in generative terms, would involve the application of the soft 
mutation rule 1.1. a to an underlying word-final voiceless aspirated 
stop. There is a major difference, however, between this application 
of rule 1.1. a and other applications of the same rule in the environments 
of the soft mutation, besides, of course, the evident difference that 
this application is in word-final position. The difference is that 
there is contrast maintained in soft mutation due to its nature as a 
morphophonological rule (compare section 5.0) » but there is no functional 
contrast in this application of rule 1.1. a, for there is no occurrence 
of a word »tec Vtek/ with which teg /teg/ may be contrasted. Not only 
is this the case with this particular adjective, but this is the case 
with all adjectives. 

It appears, then, as though we have a rule voicing and deaspirating 
word-final stops in adjectives which serves no contrastive or distinctive 
function. But the soft mutation rules apply morphophonologically in 
order to distinguish grammatical conditions. Insofar as this rule does 
not serve to reflect the grammatical conditions but applies in all 
occurrences in word-final position, what we have is not soft mutation 
rule 1.1, a at all, but the minor rule 3 A. 

The process described in adjectives can also be found in certain 
noun-verb categories. For example, the verb bwyta /buiita/ 'eat' 
alternates with the noun bwyd /but id/ 'food', the verb pysgota /pasgota/ 
'fish' alternates with the noun pysgod /pasgod/, etc. 

At this point, an historical Gymrist would probably take issue 
with my analysis. As I point out in section 2.1, the alternation 


known as 'hard mutation' is a result of an /h/ in Middle Welsh occurring 
between the adjective, ending in a voiced unaspirated stop, and some of 
the adjective endings. Moreover, the voiceless aspirated stops found 
in the verbs cited above and others like them result from the occurrence 
of an /h/ in a factitive ending applied to the noun (ending in a voiced 
unaspirated stop) also in the Middle Welsh period. In both cases, a 
voiced unaspirated stop followed by an /h/ was realized as its voiceless 
aspirated cognate. (I return to this type of realization in Chapter 7.) 

This objection rests upon a fact which was true enough in the 
grammar of Middle Welsh. In Modern Welsh, however, the phonology is 
quite different. First of all, the alternations must be described 
with relation to the data available to the modern speaker of Welsh. 
Moreover, there is some evidence that some speakers have indeed estab- 
lished phonologies which reflect this synchronic analysis. As Watkins 
(1961 j67~citing Fynes-Glinton 1913) points out, the adjective sad /sad/ 
•firm', a modern borrowing from English, has the equative, comparative, 
and superlative forms sated /sated/, satach /satax/, and sataf /satav/, 
respectively. 1 

There is one area of Middle Welsh which is of considerable impor- 
tance to the nature of Modern Welsh. In Middle Welsh manuscripts, we 
do find occurrences of word-final £, t, and c (or k). At first glance, 
we should see two possibilities in the development of word-final stops. 
Either these stops were indeed voiceless aspirated stops or they were 
orthographic devices to represent an archiphoneme (in the sense of 
Trubetekoy I969). 

If the first possibility is true (which is doubtful), then the 
minor rule 3.4 applies extensively in Modern Welsh. In almost every 


case, where Middle Welsh orthography maintains a £, t, ore (or k) 
in word-final position, Modern Welsh maintains a b /b/, d /d/, or £ /g/, 
respectively. In the few cases in which Modern Welsh orthography main- 
tains the symbol for the voiceless aspirated stop in final position in 
words inherited from Middle Welsh, the stops occur in clusters with a 
sonorant, and in the colloquial dialects (excluding Standard North Welsh, 
Standard South Welsh, and of course, the Standard Welsh) the stops there 
are seldom pronounced at all. For example, the Standard phrase written 
maent hwy 'they are' is usually rendered /maiin nhu(i)/ in speech. 

The second possibility, that the Middle Welsh orthographic symbols 
represent archiphonemes, is the more probable explanation. According 
to Watkins (I96I 179) , Middle Welsh orthography used p_, t , and c (or k) 
in word-final position to represent /b/, /d/, and /g/. Phonologically, 
there was no contrast ( cyferbyniad ) between the two classes of stops 
in word-final position. The implications of word-final position being 
a position of neutralization in Middle Welsh are quite important in our 
description of the consonant shift. The words inherited by Middle Welsh 
or borrowed during this period may have had word-final voiceless aspirated 
stops, but the nature of the archiphonemic neutralization of voice and 
aspiration in word-final stops would insure that voiceless aspirated stops 
in the inherited vocabulary and in borrowings would change to voiced 
unaspirated stops, in effect applying minor rule 3 A throughout (as I 
point out below). Thus, through the development of word-final position 
as a position of neutralization in Middle Welsh, Modern Welsh actually 
began with the effect of the proposed consonant shift already firmly 
established in word-final position. 

No matter which explanation we choose for the presence of word-final 


symbols for the voiceless aspirated stops in Middle Welsh and their 
absence from Modern Welsh, the implications for the suggested consonant 
shift are the same. Modern Welsh began with word-final position failing 
to maintain contrast between voiceless aspirated and voiced unaspirated 
stops. The question remains, however, as to what has happened in word- 
final position since that time. 

As we turn our attention to words which have come into Modern Welsh, 
we see that the neutralization process found in Middle Welsh has been 
(at least for a time) productive in the Modern Welsh period. This is 
particularly the case among the dentals and velars. For example, English 
velvet /velvet/ has been borrowed as melfed /melved/, racquet /raeket/ 
as raced /raked/, rocket /roket/ as roced /roked/, frock /frak/ as 
ffrog /frog/, and desk /desk/ as desg /desg// The fact that the 
labials do not appear to be affected does not diverge from the shift 
hypothesis, for, as we see in section 4.1, the generalization of a minor 
rule can follow a progression in time with respect to each position of 
articulation separately. 

The fact that we have any application or generalization of the minor 
rule 3.4 (or at least, through neutralization, the effect of minor rule 
3.4) in the Modern Welsh period can but converge on the notion that there 
is a sound change in this particular position in the word. To be sure, 
however, there are a few instances in which voiceless aspirated stops 
occur in word-final position. In some of these instnaces, the voiceless 
aspirated stop isfbund in free variation with a voiced unaspirated 
cognate. For example, the term celc /kelk/ 'concealment' alternates 
noncontrastively with celg /kelg/ ( celc is borrowed from Irish), although 
the term is also realized with a suffix as celciad /kelkyad/ and as a 


verb celclo /kelkyo/. Moreover, ffrog /frog/, mentioned above, can also 
be found as ffroc /frok/. 

On the other hand, there are a few instances in Standard Welsh, 
in which words maintain the voiceless aspirated-voiced unaspirated 
contrast in word-final stops. For example, the English cab /c»b/ or 
cabin /caebin/ has been borrowed by Welsh as caban /kaban/, with the 
truncated form cab /kab/j while English cap /caep/ has been borrowed 
as cap /kap/ (also with the longer form capan /kapan/). The maintenance 
of contrast in word-final stops, however, is the exception rather than 
the rule.-' 

The issue of the maintenance of contrast in word-final stops brings 
up another problem — that of bilingualism. The chief source of new words 
in Welsh is English, as most of the Welsh-speakers also speak English, 
and English has been imposed upon the country especially through the 
education system (see W.R. Jones 1966), It could well be, then, that 
it is the speaker's knowledge of English which leads this speaker to 
maintain contrast in such pairs as cab /kab/ and cap /kap/, above. 
According to Professor T. Arwyn Watkins (personal communication), 
•the phonological system of Welsh is nonproductive in Welsh, insofar 
as borrowings are concerned'. 

This interrelationship between the maintenance of contrast in 
a supposed position of neutralization and the speaker's knowledge of 
English can be illustrated further by noting the phenomena associated 
with lenia initials in Welsh borrowings (see Griff en In press a). When 
a word is borrowed from English with an initial consonant which cor- 
responds to one of those segments in the mutation system which occur 
as soft mutation forms but not as radicals (see Tables l.O.a-b), this 


segment eventually is changed to the corresponding radical form. Thus, 
we would expect initial /r/ eventually to be realized as /*/ in environ- 
ment I, As we see above, however, the terms raced /raked/ and roced 
/roked/, being fairly recent, have not yet undergone this change. None- 
theless, the final /t/ of the English words has become /d/ in the Welsh 
words. Now in these two words, the final consonant does not serve a 
distinctive function through its voice and unaspiration. In the term 
rhac /sak/ 'rack' (from the English), on the other hand, the initial 
segment has undergone the change to the radical but the voiceless aspirated 
stop is still maintained in word-final position probably because of the 
existence of a word rhag /aeag/ 'lest'. Although there is certainly no 
prohibition in the Welsh language against homophony, I would contend 
that contrast is preserved through the speaker's knowledge of the 
English word and the English phonological system. 

In this section, I have dealt at some length with the notion of 
archlphonemic neutralization of contrast. In spite of my attempt to 
incorporate the notions of contrast and neutralization of contrast into 
the generative model (see Griff en 197^) » these notions are not a part 
of generative phonology as it is presented in Chomsky and Halle 1968 
or as it is usually practiced. In the current generative model, contrast 
has largely given way completely to the notion of distinctiveness, which 
operates in the systematic-phonemic realm of the phonological component. 
The manner in which we incorporate the effects of neutralization (as 
mentioned above) is through a rewrite rule carried out in the phonological 

The rewrite rule which we would have to posit in the phonology in 
order to accommodate the phenomena noted in this section would change 


all voiceless aspirated stops to their voiced unaspirated cognates. 
In other words, we would apply minor rule 3. 4. Thus, insofar as our 
generative description is concerned (in which neutralization is handled 
by rewrite rule), the minor rule has indeed spread to word-final position. 
Moreover, as we see from most of the borrowings from the Modern Welsh 
period, the minor rule is in the process of generalizing. 

5.2 Medial Positions and Clusters . As we direct our attention to 
intervocalic position, again we should consider the type of phonological 
strings inherited by Modern Welsh. As I mention in section 1.0, 
Brythonic and Latin voiceless stops were realized in the Old Welsh 
period as their voiced unaspirated cognates (Morris Jones 1913«l6l| 
Jackson 1953*553-7). 

Although the intervocalic voicing rule is in fact the impetus 
(along with the loss of final syllables) for the development of the soft 
mutation rule, the environment for the application of this rule in Old 
Welsh is by no means limited to those environments which would develop 
into grammatically motivated soft mutation. To the contrary, the rule 
applies in Old Welsh purely on a phonological basis. Thus, we could not 
say that such a rule is parallel to soft mutation rule 1.1. a. Instead, 
it is identical to minor rule 3.4 (insofar as its application to voiceless 
aspirated stops is concerned). 

As Modern Welsh inherited its vocabulary ultimately from Old Welsh, 
the impact of the Old Welsh rule upon Modern Welsh lies in the lexicon. 
Just as the Middle Welsh position of neutralization in word-final 
position (section 5.1) created a state in which Modern Welsh would have 
few if any word-final voiceless aspirated stops at the beginning of the 


Modern Welsh period, the Old Welsh intervocalic voicing rule created 
a state in which both Middle Welsh and, through it, Modern Welsh would 
have few intervocalic voiceless aspirated stops. 

Of course, as far as the actual progression of a shift in Modern 
Welsh is concerned, such phenomena as the Middle Welsh word-final 
neutralization and the Old Welsh intervocalic voicing do not converge 
upon the shift hypothesis directly. They do, however, converge indirectly 
by helping to create the effect of the shift. Such rules, then, serve 
to establish a conspiracy— a functional unity among several different 
rules (see Kisseberth 19?0ai The fact that such a conspiracy exists 
over an extended period of time can only support the notion that a shift 
is taking place. 

Although Modern Welsh does not have an intervocalic voicing rule, 
some intervocalic voicing is nonetheless evident. The Standard Welsh 
'hard mutation' phenomenon requires that an adjective (with an under- 
lying final voiceless aspirated stop) such as teg /teg/ (cited above) 
have its voiceless aspirated stop realized unchanged in intervocalic 
position, as in the equative teced /teked/. 

There are some signs, however, that this system is changing. 
Mrs. Mona Pringle, of Gainesville, Florida, a native speaker of the 
Colwyn Bay area dialect of North Welsh, gives the equative form of teg 
/teg/ as teged /teged/. Now this is not Standard Welsh, but North Welsh 
is considered to be fairly conservative otherwise, and Mrs. Pringle does 
not, for instance, accept /3/ as an alternative form in soft mutation 
of /c/ (see Griff en 197^b). 

Moreover, as I mention in the previous section, while sad /sad/ 
'firm' was noted as having the equative form sated /sated/ in data 


collected early in this century, It is now more commonly heard as 
sadied /sadyed/. Likewise, other English loanwords seem to be affected 
in the same way. For example ffeind /faind/ 'fine' has the equatlve 
form ffelndled /foindyed/ (Watkins 196lt67). To be sure, this may be 
another case of interference through bilingualism, but the effect is 

The loss of the 'hard mutation* alternation, as it affects col- 
loquial Welsh, can certainly be attributed to analogy, from the fact 
that there is a voiced unaspirated stop in final position of the word, 
and hence in final position of the morpheme. The presence of such 
analogical leveling in a system which tends to voice final stops can 
be seen as further evidence of a tendency to replace medial voiceless 
aspirated stops with their voiced unaspirated cognates, at least where 
morpheme boundaries may be involved (as in the addition of adjective 

As we turn our attention to clusters, we find a wealth of corroborating 
evidence for such a shift occurring in Modern Welsh, According to 
Fynes-Clinton (l913txxiii), Old and Middle Welsh combinations of voice- 
less aspirated stops (in pairs) are realized in Modern Welsh in one of 
two ways. If the second element of the stop cluster is a labial or a 
velar, this segment is voiced and unaspirated in Modern Welsh. If the 
second stop is a dental, it is realized as a voiceless unaspirated stop. 
In narrow phonetic terms, then, [pt] becomes [pd], [pk] becomes [pg], 
£tp~] becomes [tb], [tk~] becomes [tg~],[kp] becomes [kb], and [kt] becomes 
[kd"] (where the symbol for a voiceless stop represents voiceless 
aspirated, the symbol for a voiced stop represents voiced unaspirated, 
and the symbol for a voiced stop with a dot subscript represents 


voiceless unaspirated). 

Moreover, when a voiceless fricative precedes a voiceless as- 
pirated stop in Old or Middle Welsh (Pynes-Glinton 1913*xxiii), the 
stop is realized as it is in stop clusters, above. If, however, the 
fricative is dental /9/, the dental stop is not only unaspirated but 
also voiced. According to Bwrdd Gwybodau Geltaidd (1942:53)» though, 
the dental stop is also voiced and unaspirated after the voiceless velar 
fricative /x/. Thus, the inheritance of Modern Welsh (at least from 
Early Modern Welsh) is such that there has been extensive voicing and 
even greater deaspiration in consonant clusters. As these processes 
have been carried out on a purely phonological basis, we can see here 
the further application of minor rule 3.4 in this new environment. 

The productivity of the minor rule in consonant clusters can be 
seen in Modern Welsh borrowings from English. For example, English 
desk /dtsk/ has been borrowed as /desg/, and English splendor /splendar/ 
has been borrowed as ysblander /(e)sblander/. 

In order to add data on the productivity of minor rule 3.4 in 
consonant clusters in contemporary Welsh and determine its extent more 
accurately, I constructed an experiment using, among other items, a 
list of words found in Table 5. 2. a. These were written in the standard 
orthography — the only difference between this and the phonological 
notation used herein is that c is used for /k/. Each was presented 
to an informant on a three-by-flve card, and the informant read each 
word once slowly, syllable-by-syllable, and once quickly, as a single 
word. The responses were recorded on a Uher 4000 Report L. 

Two informants are from the Bangor district studied in Fynes- 
Glinton 1913. Dr. Bedwyr L. Jones, Professor of Welsh at the University 


College of North Wales (Bangor), comes from northern Anglesey (Mon), 
and Mr. Idrls Roberts, on the staff of the same college, comes from 
Penrhyn, on the eastern outskirts of Bangor. Mr. Hugh Jones, a farmer 
from Bettws Garmon, speaks the Caernarvon district dialect. Mr. Merfyn 
Morgan, a researcher in the Department of Welsh at the University College 
of North Wales, cames from Newcastle-Emlyn in an area of Dyfed formerly 
known as Carmarthen, Dr. Jones was born in 1933 1 Mr. Roberts around 1914, 
Mr. Jones in 1902, and Mr. Morgan in 1947. In each case, the informant's 
parents were raised in the same region and taught their children Welsh 
as the primary language. Each informant feels that his speech represents 
the speech of his dialect area, each is bilingual in Welsh and English, 
and none was observed to have any speech impediments. 

All Informants could pronounce the voiceless aspirated stops and 
the voiced unaspirated stops intervocalically in various positions in 
the word. When the first three informants, those speaking North Welsh, 
recited the items in Table 5, 2. a, none had any trouble with them, al- 
though they all recognized that some combinations were not found in 

With Mr. Morgan, on the other hand, I found a different situation. 
He recited the list without any changes when reading syllable-by-syl- 
lable. But when he recited the items quickly as words, the result was 
the list found in Table 5.2.b. 

These results show a marked influence of the minor rule 3.4. It 
is not surprising to note that Mr. Morgan is from the same general 
region as the authors of Bowen and Rhys Jones I960 (see section 4.1). 

There is one troublesome point in the cluster phenomena treated 
in this section. Up to this point, we have only been concerned with 

Table 5. 2. a 
Standard Orthography 
















Table 5.2.b 
Pronunciation by Merfyn Morgan (Newcastle-Emlyn) 
Narrow Transcript ion 



































voiceless aspirated stops and voiced unaspirated stops. Now, however, 
we find voiceless unaspirated stops (tenues). This new class of tenues 
does not serve any distinctive function, so it can be relegated to the 
systematic phonetics as an allophone, or combinatory variant. As such, 
then, it must be derived from a phonological segment — either from the 
aspiratae or from the mediae. It is quite crucial to the arguments 
concerning us here that we determine whether the tenuis is a subclass 
of the aspirata or it is a subclass of the media, for if it is a sub- 
class of the media, then the systematic phoneraics of the phonological 
component must first derive these mediae from aspiratae — further evidence 
of the minor rule 3,k, 

Now the tenuis differs from the aspirata in aspiration and from 
the media in voice. In order to show from which class this subclass 
is derived, it is therefore necessary first to determine whether it is 
the aspiration or the voice which serves the distinctive function in 
the phonology. As I demonstrate in detail in Chapter 7, it is the 
aspiration which is functionally distinctive (contrastive) in Welsh. 
Thus, the phonologically pertinent change in minor rule 3.^ is the 
change from aspirated to unaspirated; so in the case of the voiceless 
unaspirated stops (tenues), the phonologically pertinent aspects of 

the minor rule have indeed applied, and any occurrence of such a stop 

is evidence of the spreading of the minor rule. 

Thus, the extent of application of minor rule 3.4 from Old Welsh 

to contemporary Modern Welsh in medial position and in consonant clusters 

can be seen to be fairly great. On the other hand, there are also some 

counter-processes working against the spread of minor rule 3. 4. These 

are collectively known as 'provection' (see Morris Jones 1913il8l-9; 


Jackson 1953»56l-5). In Chapter 9» I treat some of the phenomena 
classified under provectlon in an hierarchical framework. In the 
generative framework at this point, however, it is enough to say that 
provection does not block the application of minor rule 3.*M so, in the 
light of the requisites for proof given in section 5.0, the minor rule 
is indeed spreading in this environment. 

5.3 Initial Position . The first evidence for the proposed consonant 
shift in initial position is that developed in Chapters 3 and *»-. In 
the case of the deviation in the mutation system represented by gan 
/gan/, I ought to stress one important fact. Although the minor rule 
3.k has indeed brought about an historical change in the language in 
the case of the deviation, the change has not affected the underlying 
representation of the words affected. 

In the generative model, the fact that the underlying form of, 
for example, gan /gan/ is Ikanl and the fact that this underlying form 
is the same as that extant in the description of the language before 
the change demonstrate that we have here not a change in lexical item, 
but a change in the phonological rules employed in the grammar (the 
description of the language). Such a change not reflected in the lexicon 
creates a dilemma. As the underlying segment must be a |k| in order to 
derive the mutation variants in the appropriate environments, on this 
underlying level there must be a |k| distinct from |g|. On the other 
hand, there is no functional distinction in the affected words between 
/k/ and /g/ (and likewise for the labials and dentals). 

This property by which distinctiveness is maintained on the 
underlying level but absent on the phonological level is known as 


neutralization. Insofar as the neutralization in the affected classes 
of words takes place in all environments, what we have is absolute 
neutralization, which I mention in section 3.3 (see also Kiparsky 1973). 

Taking this absolute neutralization and the existence of the 
underlying voiceless aspirated stop into consideration, we might well 
ask just what the criteria are for determining a consonant shift. Is 
there to be a loss of distinctiveness in the underlying level which 
motivates the phonology and produces the acoustic data, or is a change 
in the acoustic data which reflects loss of functional contrast sufficient? 
On the one hand, the notion that phonological change is a reflection 
of sound (acoustic) change (as I point out in sections 4,0 and 5.0) 
would tend to support the latter condition? while, on the other hand, 
the fact that phonological (though functional) contrast (as opposed to 
underlying distinctiveness) Is not accepted in most approaches to 
generative phonological theory would tend to discredit the latter 
condition, supporting indirectly the former. 

Due to the formal nature of generative phonology, the question of 
the effect of neutralization on the type of change reflected by the 
deviation is moot. As I state in section 5.0, only two conditions are 
needed to show that a shift is progressing — the existence of a minor 
rule and the generalization of the minor rule. It is sufficient there- 
fore to point out that the minor rule exists in this environment and 
that it is generalizing in this environment to more and more items. 
This I have demonstrated in Chapters 3 and *K The question remains, 
of course, as to whether such a model which relies upon the formal 
device of the rewrite rule can adequately predict the existence of an 
on-going consonant shift. " I return to this question in Chapter 6. 


The type of change outlined above in which the underlying segment 
remains intact while the phonological rules change is just one type of 
change that can occur in initial position (see section 4.0). This type 
of change I label the gan-type change, for the sake of convenience. 

The gan -type change has an antithesis, in which the underlying 
segment changes while the phonological rules remain intact. This type 
of change is known as relexicalizatlon (the restructuring of section 
k,0 carried out in a particular lexical item). In relexicalizatlon, 
the minor rule applies, and the speaker, hearing the deviant form in 
environment I, reclassifies the output of the minor rule as the lexical 
item itself. In order for relexicalization to converge on the shift 
hypothesis, it must be carried out such that a segment which was formerly 
a voiceless aspirated stop has changed to a voiced unasplrated stop in 
initial position. 

As the gan -type change affects the rather limited class of function 
words, we might expect any relexicalization to affect the unlimited 
class of content words. As far as Standard Welsh is concerned, there 
have been very few changes at all in word-initial position among consonants 
since Middle Welsh, In the colloquial dialects, however, some changes 
have taken place parallel to the minor rule 3 >^> 

According to Professor Herbert Pilch (personal communication), 
in the dialect of Cardigan (now part of Dyfed), some words beginning in 
Standard Welsh with voiceless aspirated stops and taking the mutations 
of voiceless aspirated stops are found in the unmarked environment with 
the corresponding voiced unaspirated stops and the mutations of the 
voiced unaspirated stops. As an example, Professor Pilch cites the 
word pobl /pobl/ 'people'. In Standard Welsh, the word pobl /pobl/ 


alternates in soft mutation with bob! /bobl/ In this dialect, however, 
the word is bobl /bobl/ (pronounced [bobol] after epenthesis), and it 
has the soft mutation form fobl /vobl/. I refer to such a change as 
a pobl -type change, for the sake of convenience. 

The first thing that should strike us as we consider the pobl- type 
change in the Cardigan dialect is the fact that this change is not very 
extensive — affecting only a few words in one dialect and rare elsewhere. 
Nonetheless, in order for the relexicalization to take place, the voice- 
less aspirated stops had to become voiced unaspirated stops. Thus, the 
minor rule 3.4 has indeed applied. Inasmuch as the rule did not pre- 
viously apply in earlier forms of the language, the application of the 
minor rule can be seen as a generalization from previous environments. 
The same phenomenon can be seen in the Bangor dialect, as described 
in Fynes-Clinton 1913 • For example, cingroen /kir)gro:in/ 'stinkhorn' 
has the alternative forms [kiggron] and [giggron], and piogen /pyogen/ 
'magpie* has the alternative forms [pyogan] and [byogan]. 

There is an example of the pobl -type change in Standard Welsh. 
According to Fowkes 1949 (and Fynes-Clinton 1913) » the English word 
gooseberries had been borrowed into Welsh as cwsberins /kusberins/. 
According to Evans and Thomas 1968, however, the word is now gwsberys 
/gusberis/~a clear application of minor rule 3.4 and capable of being 
described as relexicalization. In the light of the effects of bllln- 
guallsm on the Welsh language (see section 5.1). attributing such a 
change to a shift in the language may be going a bit too far. 

The evidence for the pobl -type change being an integral part of 
a consonant shift in progress in Modern Welsh is, indeed, weak. On 
the other hand, we have a few cases of relexicalization that parallel 


the minor rule 3.4. We add to this a change (though it may well be 
widespread) which is indicative of the minor rule but which is more 
indicative of the effects of an ever-growing influence from English. 
We may indeed ask whether such evidence is even admissible. 

The reason for the uncertainty over the admissibility of the 
pool -type changes in Modern Welsh as evidence for a consonant shift 
lies in the area of purpose. Although the two requisites for deter- 
mining the existence of a consonant shift are both met with regard to 
the pobl- type change, it nonetheless appears as though these occurrences 
have nothing to do with the shift, but reflect other conditions in the 
language (including the 'conditions' of random change). This matter is 
treated further in Chapter 6. 

The final piece of evidence for a consonant shift in progress in 
Modern Welsh can be found in the dialect of Tf Ddewi in the area of 
Dyfed formerly known as Pembroke. According to R.O. Jones (1971»169), 
the nasal mutation in the dialect of Ty Ddewi has been lost over the 
course of three generations. In the older generation of Welsh-speakers 
in Ty Ddewi, we find the nasal mutation applying as it is in Standard 
Welsh (see Tables 1.0. a-b). The middle generation, however, has lost 
the voiceless aspirated nasals from the mutation system and has replaced 
them with their voiced unaspirated cognates. This change is by no means 
unique to the speakers of the Ty Ddewi dialect. The most interesting 
development tfckes place in the speech of the younger generation. These 
speakers have completely lost the nasal mutation, so that /b/, /d/, and 
/g/ are realized in the nasal mutation environment III as the radicals 
/b/, /d/, and /g/, respectively. However, /p/, /t/, and /k/ are 
realized in environment III not as the radicals, but as /b/, /d/, and 


/g/, respectively. 

Now as far as the voiceless aspirated stops are concerned, we 
could say that for the younger generation of speakers environment III 
has merged with environment II, and the realization of the voiced 
unaspirated stops is a result of the soft mutation rule 1.1. a. Given, 
however, the totality of the environments, according to which a given 
item must be in one of the mutation environments at any time, as shown 
in section 2.8, we cannot accept such a solution. Even in those cases 
in which two environments appear to coexist such that soft mutation 
applies to some segments and spirant mutation to others, as in section 
2.6, a 'bleeding' relationship can be established between the applica- 
tion of the rules, and such a relationship cannot be established between 
environment I and environment II in this case. 

In the generative approach to historical linguistics, if a rule 
is lost from the grammar and if that rule derived a surface phonological 
segment from an underlying segment, then the underlying segment must 
be realized (compare King 1969i46-51| Kiparsky 1971 »627-30). Thus, if 
the nasal mutation rule 1.2. a is lost in the Ty Ddewi dialect, then the 
voiceless aspirated stops must appear on the surface. They do not. 
Thus, it is necessary for us to posit a rule turning the underlying 
voiceless aspirated stops into their voiced unaspirated cognates. Such 
a rule is minor rule J.k. 

At least from the generative standpoint, this last piece of 
evidence is quite compelling. It is an indication that, in a position 
of instability where a rule is lost requiring the surfacing of the 
underlying segment, the minor rule intervenes. The implication of such 
a development is that any addition to environment I is going to motivate 


the change exemplified by minor rule 3 #4. Considering the nature of 
the mutation system environments as outlined in section 2,8, such a 
development as that mentioned above can only give impetus to a consonant 

In this section, I have drawn evidence from three phenomena in 
word-initial position for the concept of a consonant shift developing 
around minor rule 3.4. Although the notion of the pobl -type change 
(relexicalization) may not be very strong, the evidence afforded by 
the gan-type change and by rule loss in the Ty Ddewi dialect is quite 
strong from a generative point of view. Any evidence with the least 
amount of credibility, moreover, is especially persuasive if it occurs 
in word-initial position. The reason for this is the stabilizing 
influence of the mutation system. In order to maintain grammatical 
information within this system of initial consonant gradation, contrast 
(if not distinctiveness) must also be maintained between the voiceless 
aspirated stops and the voiced unaspirated stops. 

5.4 Quod Erat Demonstrandum . In this chapter, I have set out to prove 
that a consonant shift is progressing in Modern Welsh, In order to 
prove this, two conditions should have been met — to show that the minor 
rule 3.4, a single rule changing an entire natural class, is a necessary 
part of a description of Modern Welsh, and to show that minor rule 3.4 
has indeed been generalizing in different environments over a period 
of time, 

I have presented the evidence in three parts, paralleling the 
three positions within the word. In final position, we find a position 
of neutralization which historically has acted in accordance with 


minor rule J.k to change all voiceless aspirated stops to voiced 
unaspirated stops. As we see from some fairly recent borrowings, more- 
over, where the effects of bilingualism are minimal, the minor rule is 
productive in this position. Thus, the minor rule exists and is generali- 
zing in word-final position. 

In medial position, we find that borrowings from English which 
at one time adhered to the 'hard mutation' system no longer do so, A 
change can further be seen in at least one dialect in which the voice- 
less aspirated stops in medial position in adjectives have changed 
likewise to voiced unaspirated stops. Furthermore, in clusters, we 
find a large number of instances in which historically voiceless aspirated 
stops have become voiced unaspirated stops phonologically (although 
some become voiceless unaspirated stops phonetically). Such changes 
are not only evident in the transition from Middle Welsh to Modern Welsh, 
but are also found in Modern Welsh dialects. Thus, in medial position 
and in clusters, the minor rule exists and is generalizing. 

Finally, in initial position, we find the gan-type change which 
(as shown in section 4.1) is generalizing today. The pobl -type change, 
moreover, shows the evidence of the minor rule 3.^ and is a factor 
which constantly changes. In the loss of the nasal mutation rule in 
Ty Ddewi, we find the application of the minor rule generalizing in the 
younger generation of speakers. Thus, in initial position, too, the 
minor rule exists and is generalizing. 

It would appear, then, as though the conditions for a consonant 
shift along the lines of the minor rule J.k with an unrestricted 
environment have indeed been met. Insofar as these requisites are both 
necessary and sufficient in the generative model to prove the existence 


of an on -going consonant shift, I should submit that the shift is 

Yet, in spite of the fact that I have proven the existence of a 
consonant shift, I have not proven that there is a change occurring in 
the Welsh language. All that I have shown is that a consonant shift is 
in progress in the generative description of Modern Welsh. If the 
generative description of Modern Welsh is indeed adequate in all respects, 
then we can say that there is, in fact, a shift occurring in the language 
Itself. I address the adequacy of the generative description of Modern 
Welsh in the next chapter. 


*To be sure, these are more usually heard as sadled /sadyed/, 
etc., today, a point to which I return in the next section. The 
word is borrowed from Scot and Northern English dialects, as in the 
term 'sad bread* (Fynes -Clinton 1913 •W-l). 

2 I return to this particular problem in the next section. 

3professor Herbert Pilch (personal communication) offered only 
two for the colloquial dialect of Cardigan (now Dyfed)-- twp /tup/ 
•stupid' and gwep /gwep/ 'grimace'. 

Within the framework of generative phonology, such an argument 
is vague. This vagueness, however, should be cleared up in Chapter 7. 

5This is not to claim that other phonological theories without 
the rewrite rule can achieve this goal either. 



6.0 Problems In the Generative Description . As pointed out In section 
5,0, actual sound change ought not to be confused with the conventionalized 
description of sound change, be it In the generative model or in any 
other model of linguistic description. There is always a danger in a 
linguistic description of confusing the logical notion of validity with 
actual reliability. 

Insofar as the description and arguments of the past five chapters 
are Internally consistent and they relate the premises (the aspects or 
hypotheses of the generative framework) to the observations about the 
Welsh language, our generative description is quite consistent. But 
is it reliable? This is important, for if the generative description 
is not reliable, then we cannot infer from the consistence of the 
framework that there Is even such a thing as the new Welsh consonant 
shift. Without establishing the reliability of the framework, then, 
we cannot conclude anything about the language being described—only 
about the framework being used. 

In order to establish the reliability of the framework in the 
description of Welsh in general and the consonant shift in particular, 
we must demonstrate that those aspects of the framework which directly 
(or, indeed, indirectly) affect our proof are, in fact, adequate. As 
the reliability of the system depends upon the total affect of all 
pertinent aspects of the system, a single inadequacy will result in 



the reliability of the framework, insofar as the shift is concerned, 
being discounted* In such an event , the conclusions of the description 
may very well stand, but the framework mist be changed before those 
conclusions can be drawn from the generative model. 

Considering the importance of assuring reliability in any description 
of a language event, we should at this point look back upon the descrip- 
tion and arguments presented in the generative model in order to determine 
whether there might be any inadequacies which could endanger the con- 
clusions reached in Chapter 5. Of course, any such inadequacies must 
be corrected either within the generative model or in some subsequent 
model of linguistic description. 

The first inadequacy which I should like to address concerns one 
of the most basic notions of a generative description — a generative 
description must be formal and explicit (see Lyons I9681I36-7). In 
itself, such a requirement is hardly objectionable, but if the descrip- 
tion relies upon notions which are not formal and explicit, then there 
is not only an inadequacy in the description, but an inconsistency. 

The explicit formalism of generative phonology in the description 
of historical linguistics is limited to the rewrite rule. The estab- 
lishment of the consonant shift in the generative description, as out- 
lined in the previous two chapters, is predicated upon the development 
and generalization of a minor rule changing one natural class of segments 
into another natural class. Indeed, the change itself can be described 
as a change from one rewrite rule to another. 

Now the question becomes 'How do we check the reliability of these 
rules?' According to Chomsky (19651I2), the output of the grammar is 
judged in accordance with • global • considerations. Although it is 


nowhere stated what these 'global* considerations are, I would assume 
that they consist of several conventions. The first convention is the 
specification of the environment of the rule such that it applies only 
to those segments which are to be affected. When two or more rules 
could be applied in a given situation, however, we must rely on notions 
of rule ordering. Such conventions as the noniterative rule convention, 
transitivity of application, and asymmetrical application of rules are 
used in Chomsky and Halle 1968 in order to maintain linear ordering 
(even within instances of cyclical ordering), so that the output of 
one rule is the input of the next in an orderly progression. Other 
devices include the notion of maximal application (Klparsky 1968) , 
local ordering (S. Anderson 1969)1 and derivational constraints (Mls- 
seberth 1970b \ Shibatani 1973). 

The minor rule, as it is developed in Chapters 3, k, and 5, 
adheres to all of the pertinent conventions of phonological description 
in the generative model. As such, it is consistent with the system. 
However, can such a notion as the minor rule, consistent as it may be 
with the rest of the description, be reliable enough to support our 

The minor rule may well be too powerful due to the requirement of 
a formal and explicit description. This can be illustrated with the 
following hypothetical example. Suppose that in Old Welsh, nine words 
changed such that each member of a natural class of voiceless aspirated 
stops became voiced and unaspirated. Suppose further that the same 
thing happened in Middle Welsh and the same in Modern Welsh. If all 
we have to establish the existence of a consonant shift in Welsh is 
the development and generalization of a minor rule, then we would have 


to conclude that a shift Is in progress. Of course, any linguist 
who would suggest a consonant shift on the basis of so few forms would 
be ridiculed. But here we are relying upon the judgment of the linguist. 
If we are to maintain in practice the requirement that the description 
be formal and explicit, then we must supply a formal and explicit device 
for comparing one set of rewrite rules with the rest. So long as we 
do not have such a device and the linguist must rely upon common sense 
in the derivations of the phonological component, then the requirement 
that the description be formal and explicit is best ignored. Ignoring 
the requirement, however, is only going to establish an inconsistency 

in the generative model — an inconsistency which can be remedied only by 

either providing the necessary formalism or withdrawing the requirement. 

Another problem in the description concerns the notion of contrast. 
Distinctiveness is a given factor in the lexical items, which, with their 
distinctive phonological specifications, are presented to the phono- 
logical component for interpretation (application of phonological 
rewrite rules). In the phonological component, the lexical items become 
the underlying strings. This creates a distinct level. The phonological 
rules then apply in order to derive the phonetic data. This derivation 
is direct, with no intermediate level (for discussion, see Lockwood 1975). 

In this respect, my description over the previous five chapters 
is not at all typical of the generative school, for I have recognized 
three levels— the underlying level, the phonological level, and the 
phonetic level. In so doing, I follow Schane 1973i though that work 
is also atypical of this school in this respect (see also Griff en 197***). 

According to the tenets of generative theory, as set down in 
Chomsky and Halle 1968, there is no Intermediate level. This state was 


established as early as Halle 1959, and in an answer to Householder 
1965, Chomsky and Halle I965 attacks the notion of an intermediate 
•phonemic' level outright. In effect, the disestablishment of the 
phonemic level (which had been religiously maintained in American struc- 
turalism) meant that the system would not only be simplified, but it 
would also lose the notion of (functional) contrast within the phonology. 

The loss of the notion of contrast in generative phonology is 
important to the matter at hand, and we ought to consider the effect 
of this loss. Distinctiveness serves to differentiate segments in classes 
on the underlying level, while contrast does the same on the phonological 
level. By eliminating the phonological level of contrast, the generative 
theorist holds that simplicity is afforded at no cost, for the rewrite 
rules are fully specified and act upon feature specifications found on 
the underlying level. Thus, it is maintained, there is no need in the 
structure to add a redundant level at which contrast is reestablished 
(compare the derivation axiom in Lockwood 1975). 

In language change, however, there is definitely a need for some 
determination of contrast less abstract than the underlying level. 
In fact, this notion of contrast is central to the use of the minor 
rule in the deviation. As we see in the synchronic situation in Chapter 
3 and the diachronlc situation in Chapter k t the underlying initial 
segment of gan /gan/, for example, is |kl. This is quite evident from 
the mutation forms. After the minor rule 3.^ applies, though, the 
initial segment becomes /g/. As we look at this situation from its 
historical viewpoint, we note that the underlying segment is still 
maintaining distinctiveness. So what has changed? In the generative 
model without contrast, we must say that a rule has changed, but we 


cannot say what the effect of this change has been. Without contrast, 
we cannot interpret the change as a shift in consonants — only as an 
addition of a rule. 

Moreover, in the synchronic description, where (in section 5.2) 
cluster phenomena create a voiceless unaspirated stop (tenuis), without 
a level of contrast the rule devoicing the phonological segment applies 
at the same level of the phonology as the minor rule. This makes it 
impossible to differentiate between a change of phonological pertinence 
and one of phonetic pertinence (in the manner of the Prague school). 
In such a system, a rule in the derivation applied because of the 
physiological limitations of the articulatory apparatus is just as 
grammatically-oriented as a mutation rule. Although such a practice 
may be adequate, it is hardly appropriate for achieving insights, 

A further criticism of the system is that it lacks any kind of 
teleological principle (see Jakobson 1962 t 1-2). Such a principle 
would indicate how a language is changing or could change given certain 
conditions within the language. It would establish goals of language 
change and classify changes within the language with reference to these 
goals, (Compare, for example, Welch 1975 and the 'panchronic' approach.) 

One could object to this criticism by pointing out that the minor 
rule and its generalization describe conditions of language change and 
the goals of this change. But they do not for several reasons. First 
of all, a generative description must be formal and must be explicit. 
The informal and inexplicit description of goal stated above could not 
be incorporated into a generative grammar. Moreover, if we were to 
attempt to include such a statement of goal into a generative grammar, 
we would find that the statement may lend itself to a logical Implication 


(hypothesis), but it is unsuited to current formalism. 

As mentioned in section 5.3, there is an attempt in generative 
theory to accommodate some form of teleological principle. This is 
found in the markedness conventions of Chomsky and Halle (19681 Chapter 9). 
I treat this attempt in section 6,2, 

6.1 The Inconsistencies of Soft Mutation . One of the major probl< 
in any generative analysis of Modern Welsh lies in the treatment of 
soft mutation. As stated in Chapter 1 and Chapter 2, there are two sides 
to the mutation system — the phonological side and the grammatical side. 
Now in nasal mutation and spirant mutation, we find two finite sets of 
grammatical environments each motivating a phonological rule (or set of 
rules, if we wish to maintain a distinction between the two classes 
of stops in nasal mutation) which represents one particular phonological 
process. On the other hand, the soft mutation adds an element of 
irregularity to the otherwise regular system. 

In section 2.2, I outline the environments which (according to the 
generative approach) motivate the various soft mutation rules. Through- 
out Chapter 2, I emphasize that whenever a mutation environment exists 
and the initial segment that falls into this environment is one of those 
segments noted in Chapter 1 as undergoing the particular rule, then the 
rule must apply. Although, as we see in section 2.6, some underlying 
segments may escape the application of soft mutation due to the bleeding 
relationship found between environment IV and environment II, a segment 
which can undergo soft mutation and which is not covered by a counter- 
example or exception (section 3.0) must undergo soft mutation, regardless 
of whether this segment is supposed to undergo rule 1.1. a, 1.1. b, 1.1. c, 


1.1. d, or l.l.e. 

Soft mutation, then, is a single process, which applies in a 
specified environment. As we see in section 1,1, however, in spite of 
the fact that soft mutation is a single process, it is represented in 
the phonological side of mutation as five separate processes. I do not 
wish to contest the notion that with some extremely complex phonological 
rule, we could not use the generative notational devices (such as 
angled brackets) in such a way as to collapse the five soft mutation 
rules as they are presented in Chapter 1. I would contest any notion, 
however, that would contend that by collapsing these five rules into 
one, we have collapsed five processes into one process. 

The soft mutation rules 1.1. a-e represent the following processes i 
voicing and deaspiration; continuance) deletion i denasalization, con- 
tinuance, and desonorization t and voicing, deaspiration, vocalization, 
and sonorlzation. If one were to argue that by collapsing a rule we 
collapse the processes, then one would have to contend that, for example, 
voicing is in reality the same process as continuance, and that four of 
these processes share the same 'meta-process* as deletion. In the 
generative feature system as stated in Chomsky and Halle (l968tChapter 7), 
I do not believe that such an argument could be defended. 

In soft mutation rules l.l.d and l.l.e, moreover, we find a logical 
problem with any attempt to classify soft mutation as a single phono- 
logical process. In rule l.l.d, the affected segment takes on the 
specification [+obstruent] (desonorization) i while in rule l.l.e, the 
affected class takes on the specification [-obstruent] (sonorlzation). 
In the binary feature specification system currently used in generative 
phonology, it would be difficult to maintain that a aorphophonological 


process (neither assimilative nor dissimilative) could produce the 
opposite specifications at one time, 3 

The problem in soft nutation is that we have one grammatical 
process reflected in several phonological processes, a situation 
conspicuously absent from the other mutations. There is something 
inherently counter-intuitive about this situation. Indeed, tradi- 
tionally Welsh grammarians have considered the grammatical process of 
soft mutation to correspond to the single phonological process known as 
•lenition' (see, for example, Jackson 1953»5^3-6o). The question 
immediately comes to mind as to why generative theoreticians do not use 
such a feature. 

Generative phonology has developed from a foundation constructed 
by two principles! the phonetic feature and the binary specification. 
The features of generative phonology have been inherited from the 
American structuralist school and the Prague school. The features them- 
selves have always been based upon phonetic correlates in accordance 
with the 'inner approach' to phonology of Jakobson and Halle 1962, 
though their substance has changed from acoustic (in Jakobson, Fant, 
and Halle 1952 i Jakobson and Halle 1962) to articulatory (in Chomsky 
and Halle 1968). 

The binary nature of feature specifications was first formally 
established in Cherry, Halle, and Jakobson 1953. This system has 
maintained its basis in mathematical logic throughout its development. 
The only modifications of the basic binary system have occurred in 
Halle 1959, in which the zero specification is disestablished (and with 
it contrast— the function was replaced by morpheme structure rules), and 
in Chomsky and Halle 1968, in which notions of markedness and the M and U 


specifications are introduced (see section 6,2). 

According to Cherry, Halle, and Jakobson 1953, gradual features 
(those with degrees of presence rather than presence versus absence — 
compare the gradual versus the privative opposition of Trubetzkoy 
1969) can be described through other, binary features. This practice 
maintains the phonetic orientation of the features as well as the 
mathematical nature of the description. 

Leaving the generative approach aside for the moment, let us 
consider the fort Is lenis scale as it appears in Table 6,1 (compare 
Griff en In press a). In this scale, the lenis-most column is on the 
left, and the columns progress in strength to the right. In such a 
scale of gradual oppositions, soft nutation consists of lenition — the 
loss of one degree — in columns 2 and 3. 

Table 6,1 
The Fortis-Lenis Scale 

1 2 2 it 

v b (m) p f 

6 d t e 

g k x 

1 1 

r 9 

Unlike the generative approach, such a scale maintains a single 
phonological process parallel to a single grammatical process (soft 
mutation). Such a system of strength has been suggested in Foley 1970a 
and 1970b. The orientation of Foley's strength system, however, is not 
phonetic, but purely structural. This is to say that there is no 
phonetic process underlying the rule. What Foley has done has been to 


note the manner In which change occurs and alternations occur in 
languages and develop these into patterns. As such, then, it is 

The gradual oppositions embodied in the strength scales of Foley 
1970a and 1970b are inappropriate in generative phonology , inasmuch 
as the orientation of these scales is structural rather than phonetic. 
The 'inner approach' of Jakobson and Halle 1962 is a basic tenet of 
the theory. 

As demonstrated in Hooper (1973 'Chapter fy the gradual opposition 
along fortls-lenis lines may have phonetic correlates. Moreover, in 
the subsequent chapters in this work, I give further phonetic evidence 
for the fortls-lenis gradual opposition. The establishment of phonetic 
correlates, however, only satisfies one of the two principles in the 
foundations of generative phonology. 

Still to be accommodated is the principle of the binary nature 
of features. If we consider Table 6,1, we can see that elements of 
column 2 alternate in soft mutation with the corresponding elements of 
column 1. If this were all that is involved in soft mutation, then 
the system could be accommodated in binary features—column 2 could be 
plus and column 1 could be minus the feature [fort is]. Unfortunately, 
elements of column 3 alternate in soft mutation with the same elements 
in column 2 which alternate with elements in column 1, Such a situation 
eliminates any effective accommodation between the fortls-lenis scale 
and the binary features of generative phonology. 

Because of the impossibility of accommodating the fortis-lenis 
scale with the binary feature system, we are forced into the situation 
in which, following Cherry, Halle, and Jakobson 1953» the gradual 


opposition must be expressed through other, binary features. Thus, 
although there is certainly a single phonological process (the process 
of lenition) reflecting the single grammatical process of soft mutation; 
in the generative model, we must describe the single grammatical process 
as corresponding to five disparate and even contradictory processes. 

In the past five chapters, I have used the binary features of 
Chomsky and Halle I968 and have been able to describe all of the phono- 
logical rules, even those of soft mutation. It is not a question, then, 
of the binary system of generative phonology being adequate. The question 
is whether such a system is appropriate for the description of Velsh. 

I would contend that the generative phonological description of 
Welsh is Inappropriate, for it is contradictory. The reason that a 
phonetic orientation is used for the features of generative phonology 
is that such an orientation is natural. It maintains a link with the 
real world of observable data. Indeed, it is the appeal to this real 
world, through the findings of experimental phonetics in motor theory 
(for example, Halle and Stevens 196*0, that compels Chomsky and Halle 
1968 to adopt the articulatory substance for the features over the 

On the other hand, the requirement that the features adhere to 
some preconceived notion of two-valued mathematical logic (that is, the 
system of binary specifications) is hardly in accord with the require- 
ment that the features themselves have natural phonetic correlates. 
If it is so important that the features themselves be dictated by 
observable phonetic data, then it should be of equal importance that 
the specification of these features reflect the behavior of natural 
languages (or, at least, of the natural language being described— Welsh). 


Of course, I have been concerned here with the use of generative 
phonology for the description of Welsh. As far as theory is concerned, 
however, such a basic inconsistency as that found in the phonetic 
orientation versus the mathematic-logical nature constitutes far more 
than an inappropriateness in theory construction. It constitutes a 
flaw, saking binarity inappropriate to the object of description, and 
the nodel necessarily inadequate. 

As in the previous section, we may ask whether the Inappropriateness 
of generative phonology in this respect can be remedied. Any remedy 
that would result in a generative description of Velsh would have to 
accommodate the gradual opposition. This can be accomplished by sub- 
stituting numbers for plusses and minuses in the rewrite rules (compare 
the manner in which Schane 1973 employs numbers in the systematic 
phonetics). Such a rewrite rule could be posited as follows i 

|2 fort is) ^ fl fortis} 
l\3 fortis/i ^2 fortisJi 

The fact that the rewrite rule can be adjusted so as to accom- 
modate the gradual opposition is a positive development in generative 
phonology. Such a development, though, cannot overcome the inadequacies 
discussed in the previous section, 

6.2 Markedness and the Minor Rule . The notion of markedness is 
mentioned in section 6.0 as somehow relating to a teleological prin- 
ciple. The generative version of markedness creates a problem for the 
description presented of such magnitude that I treat it here in a 
separate section. 

To understand the generative version of markedness, it is necessary 


first to consider the Prague concept as it was developed during the 
classical Prague period (1929-1939) and as it is described in Trubetzkoy 
1969* The oppositions of the Prague school correspond to the features 
of the generative, in that they maintain a phonetic orientation and 
serve as the basis of the phonology. The Prague oppositions, however, 
are not necessarily binary, but are considered to be •strong* or 'weak' 
in any phoneme. In privative oppositions, strength is presence and 
weakness absence, but in gradual oppositions, there are degrees of 
presence, or strength. 

In any Prague opposition, strength is proportional to the deviation 
from the natural state of breathing. Hence, as breathing is voiceless, 
in the correlation (that is, the privative opposition) of voice, the 
voiced member is the strong opposition member and the voiceless member 
is the weak opposition member. Similarly, in the correlation of tension, 
as breathing is lax, the fortis member is the strong opposition member, 
and in the correlation of aspiration the strong opposition member is 
the aspirated member. 

In such a system, markedness is strongly correlated to the notion 
of naturalness (see Trubetzkoy 1969il46), If in a given phoneme the 
strong (or a stronger) opposition member is present, then that phoneme 
is marked for that opposition. If the weak (or a weaker) opposition 
member is present in a phoneme, then that phoneme is unmarked with 
respect to that opposition. Thus, a phoneme which is voiceless is 
unmarked for voice, and one which is voiced is marked for voice. To 
be sure, this situation in relation to voice is contradicted in Trubetz- 
koy (1969tl46fc but as this one reference contradicts the author's own 
notion of markedness as well as other references in his book, this 


contradiction aay be an error vhich would have been corrected had 
Trubetzkoy lived to complete the work. Whether a contradiction or 
not, however, Trubetzkoy (1969«146) goes on to note that in the final 
analysis just what is considered strong and weak in a particular opposi- 
tion (and through this, what is considered marked and unmarked) can 
only be determined by the functional structure of the particular 
phonemic system under study. 

This last point is crucial in understanding the Prague approach 
to markedness (indeed to phonology as a whole). Each language con- 
structs its own system of functional relationships, and a particular 
language can only be judged with reference to its own phonological 
system. The oppositions in one language (in the Prague approach) may 
be of an entirely different nature from the corresponding oppositions 
in another. Thus, aspiration may be a (phonetic) privative opposition 
(correlation) in English but a gradual opposition in Welsh (compare 
Kim 1965 on Korean). 

Now in historical linguistics, the notion of markedness is quite 
important. If one phoneme (in the Prague sense) changes to another 
along the lines of a particular opposition (for example, voiced to 
voiceless) which is phonologically pertinent (relevant), then we would 
expect the change to take place such that the marked oppesition member 
gives way to the unmarked opposition member. This notion hinges on the 
distinction between phonologically pertinent oppositions and phonetically 
pertinent oppositions. 

The phonologically pertinent opposition (with reference to two 
oppositions) is the opposition which can be used to predict the member 
of the phonetically pertinent opposition. Thus, those oppositions of 


phonetic pertinence only can be predicted by those of phonological 
pertinence (the 'phonemic* oppositions). The precise relationship Is 
determined by the functional structure of the phonology of the particular 

The generative approach to markedness is descendant from the Prague 
approach with some important differences. First of all, in generative 
phonology there are no notions of strength or weakness, nor of gradual 
presence. As I mention in the previous section, all feature relationships 
must be binary — implying the presence or the absence of a particular 

Not only are the relationships between feature specifications 
to be described in one particular way (binary) for all features, but 
the features themselves come from a finite set (see Chomsky and Halle 
1968sChapter 7; compare also Sampson 1974). As I mention in section 
5.0, this finite set of features is meant to be universal. A feature 
in one language is precisely the same as the corresponding feature in 
any other, at least insofar as the phonological rules are concerned — the 
systematic phonetics may realize degrees of presence. 

The notion of systematic phonemics, which must be specified using 
binary features only, versus systematic phonetics, which allows various 
degrees of realization depending upon language, may at first glance 
appear to be similar to the Prague notion of phonological pertinence 
versus phonetic pertinence. As I point out in section 6.0, however, 
whereas the rewrite rule is the sole operator and whereas there is no 
phonological level of contrast between the systematic phonemes (under- 
lying segments) and the acoustic signal, any attempt to draw a division 
between the phonology and the phonetics at all similar to the 'principium 


divislenis' of Jakobson (1962:23) must necessarily be inadequate, for 
It would contradict the basic theoretical and descriptive tenets of 
generative phonology. 

Assuming that some sort of phonological -versus -phonetic division 
could be made (as practiced by Schane 1973 and in this treatment of 
Welsh), notions of markedness would occur in the phonology (systematic 
phonemics). As such, then, a generative approach to markedness would 
be necessarily binary due to the binary nature of the systematic phonemics. 
Examination of Chomsky and Halle ( 1968s Chapter 9) reveals that it is. 

Perhaps the most crucial difference between the Prague and generative 
notions of markedness lies in the two approaches to phonological systems 
as members of the universe. On the one hand, the Prague approach holds 
that each language is judged by its own system of functional relation- 
ships. On the other hand, the generative approach maintains a universal 
base underlying all phonological systems (indeed, all linguistic systems). 

The universal base of generative phonology is reflected in its 
concept of markedness. Just as a particular feature is specified without 
regard to the operation of the phonological system of which it is a part, 
a segment is considered to be marked or unmarked on the basis of a 
universal set of interpretive conventions. These universal interpretive 
conventions determine what is marked and what is unmarked in Language. 

As in the Prague approach, the generative notion of markedness 
contends that in cases of historical change, the change of one segment 
to another reflects a change from marked to unmarked. In keeping with 
the formally explicit nature of generative phenology, the manner of 
change is not only stated in the universal interpretive conventions 
themselves, but also through 'linking rules' which serve to preserve the 


unntarkedness of the new segment. The inference to be drawn from the 
notions that narkedness is necessarily universal and that change neces- 
sarily reflects markedness is that change occurs along the lines of 
universal tendencies. 

Returning once more to the changes in the Welsh consonant system, 
whether there is an on-going consonant shift or not, there has certainly 
been a consistent (from at least one point of view) change in some cases 
from a voiceless aspirated stop to a voiced unaspirated stop. The change, 
moreover, can be represented in generative terms in the form of the 
minor rule 3.4 (at least phonologically ) . 

Inasmuch as the markedness conventions represent universal ten- 
dencies in language change and as such serve as a teleological principle 
predicting the manner of change, the change represented by minor rule 
3,4 ought to be reflected in the markedness conventions. Let us there- 
fore apply the markedness conventions of Chomsky and Halle (1968i404-7) 
to the minor rule. 

Aspiration (the heightened subglottal pressure of Chomsky and 
Halle 1968) is not covered in the markedness conventions, Indeed, it 
is usually considered to be a redundant feature. If we examine the 
specifications in Table l,0.c, moreover, we find that it is indeed 
redundant with regard to the feature voice. Thus, we can conclude that 
determining the markedness of voice must precede aspiration, which can 
be determined by some subsequent linking rule. 

Turning our attention to voice, we find that in order to determine 
the marking of voice from convention (XXI), we must first determine the 
specification of sonorance. Looking up the affected segments in Table 
l.O.c, we find that all stops are specified [+obstruent]. Thus, the 


affected segments are [-sonorant]. Applying convention (XXI) now, we 
find that the unmarked specification for voice in a segment which is 
specified [-sonorant] is [-voiced]. 

Taking the markedness information into consideration, we return 
to the minor rule and find that the segment specified [-voiced] is 
changing to [+voicedl. Thus, the unmarked segment is changing to the 
marked segment. This is contrary to the principles of markedness. 

Immediately, we are faced with a choice. Either the basic principles 
of sound change and markedness are wrong or the generative approach 
to these principles is wrong. The first choice is difficult to make, 
because the principles of markedness have served the Prague school 
(and others) so well, This leaves the second choice. 

The problem with the minor rule and the universal interpretive 
conventions has to do with the very nature of the generative approach 
to markedness. This approach maintains that the bases of the feature 
system of one language are the same as those for all languages — that 
there is a universal ideal toward which all languages 3trive, This 
can be disproven simply by noting that German has developed a system 
in which word-final stops devoice, while Welsh (as we see in section 5.1) 
has developed a system in which word-final stops voice. If both languages 
were striving toward the same goal, and if the functional relationships 
within each language's phonology were secondary to universal considera- 
tions, then such a situation in which the two languages are pulling in 
opposite directions would never come to pass. 

The problem which the consideration of word-final stop development 
In German and Welsh poses goes far beyond the adequacy of those particular 
tentative markedness conventions found in Chomsky and Halle 1968, This 


problem attacks the very notion, fundamental to generative theory, that 
markedness conventions can be constructed on a universal basis. 

The basic principles of markedness can handle this aspect of 
sound change without any problems. An example of a more realistic 
markedness convention is found in Trubetzkoy (l969«76-7). In cases of 
archiphonemic neutralization, as in cases of language change, if one 
member of an opposition is realized in the position of neutralization, 
then that member constitutes the unmarked member of the opposition in 
that particular language. If /t/ and /d/ enter into a correlation, and 
if /t/ is realized in the position of neutralization, then the phono- 
logically pertinent oppoaition ia voice, and tenaion ia merely of 
phonetic pertinence. If, on the other hand /d/ ia realized in the 
poaition of neutralization, then the phonologically pertinent oppoaition 
i8 tension, and voice i8 merely of phonetic pertinence. German, then, 
would be judged aa a language with functional phonological relationahip8 
auch that voice determinea tenaion » while Welah would be judged aa one 
in which tenaion determinea voice. Aa I demonatrate in the next chapter, 
auch a description of Welsh ia adequate. 

The Prague school provides a realistic approach to markedness 
not found in generative phonology, at least as it stands now. In order 
to incorporate such a notion as that given in Trubetzkoy (1969*76-7), 
we would have to maintain several rewrite rules, one for each of the 
variou8 language typea. Such rules, however, could not be applied 
without reference to other points in the system, for a determination 
would first have to be made as to what types of functional relationships 
characterize the particular phonology. 

Of course, we could first categorize the world's languages according 


to the functional relationships in the phonologies. Then we could 
apply the appropriate rules to each type* This would abandon the 
generative theoretical notion of universality, which forms a basic tenet 
of the theory. Indeed, without the notion of universality, 'explanatory 
adequacy % which seeks to establish the innate rules of L a n gua g e, would 
be impossible to attain using the linguistic methods of the generative 
school (thereby, according to Chomsky 1964, making all linguistic 
inquiry pointless). 

6.3 Directions for Phonological Inquiry . As we examine the description 
and argument of the first five chapters, one consideration stands out. 
Although the premises for proof may be faulty, the description exhibits 
a remarkably consistent pattern. We may not have a proof, but this 
pattern points to a convergence. And this convergence cannot be Ignored. 
It is up to the Cymrist, then, to investigate this matter more thoroughly, 
to gather and consider evidence, and to monitor whatever developments 
may occur in the passage of generations. 

Such an investigation cannot proceed, however, in a theoretical 
vacuum, for no meaningful collection and collation can be undertaken 
in the absence of premises, however tentative, that serve to direct 
the collection and collation. The construction of a more adequate 
theory of phonology is the task of the linguist. The linguist cannot 
stop with a theory of phonology, though— the theory must produce a 
model capable (at least) of describing Welsh in a consistent manner. 

Now there are countless ways of constructing a theory, but we 
need not begin with no points of reference. Indeed, there have been 
many theories of phonology, generative among them, with strong points 


and with weak points, and we can learn from both. Proa these points 
of reference, we can discern several directions for phonological 

One principle which has figured prominently in many theories is 
the 'inner approach' (Jakobson and Halle 1962). According to this 
principle, our phonology must be based upon observable data from phonetics. 
This is particularly important if we are to construct a phonology which 
can be applied to the sound system of a natural language. By beginning 
with the observable phonetics and abstracting the phonology, we insure 
that this connection between the phonology and the sound system of a 
natural language can be made. 

Another notion that should guide us in the construction of a more 
adequate theory of phonology is 'naturalism'. This is to say that the 
elements of the phonology should be constructed in such a way as to 
reflect observations of language itself, rather than in a way which 
reflects some preconceived mathematical notions. This is not the same 
as the inner approach, for we could construct a phonology from the 
phonetics using principles unsupported by the data. For example, we 
could abstract features directly into a binary mode in cases which would 
not normally support a binary mode (compare section 6.1), in order to 
achieve consistence with previously stated notions of linguistic rela- 
tionships. Of course, some hypothesizing about the structure of language 
is necessary, but this must not contradict the evidence. 

The phonology, then, should be based upon data observed in phonetics. 
Of course, it cannot be simply the phonetic data, but must be an abstrac- 
tion from them. In abstracting the phonology, each level of abstraction 
should represent a degree of pertinence which serves a particular 


function In the system of communication. That which is abstracted 
from one level of abstraction to the next must be represented on the 
lower level of abstraction. All abstraction mast be directly motivated 
by that which can be observed, not by that which can be conceived. 
In the succeeding chapters, I attempt to determine the basic 
phonetic relationships in the Welsh consonant system. These are then 
organized into a phonological system (or, more precisely, a subsystem) 
consistent with our present knowledge of phonetics. 


*This is the only formalism stated in King 1969 and Kiparsky 
1971. Of course, the linguist uses many methods of analysis, or 
•discovery procedures', in addition to the rewrite rule. Such methods, 
however, are not formally a part of the generative model. 

2This is based upon the assumption that we cannot maintain a 
logical system in which practice is not consistent with theory. 

3of course, I realize that a mechanical rule can be written in 
this way, but the notion of rewrite rule is hardly isomorphic with 
the notion of process. 



7*0 The Basic Opposition of Welsh Consonants . Acoustic and physi- 
ological oppositions (or features) have their basis in the phonetic 
data. Indeed, much effort has been expended by phoneticians in the 
last several decades in order to provide phonetic observations for the 
justification of certain oppositions as well as todetermine oppositions 
consistent with observed phonetic characteristics. For example, the 
phonetic basis of acoustic oppositions is examined throughout Fant 1973» 
and the acoustic examination of consonants has been quite extensively 
treated in the literature (for example, Delattre, Liberman, and Cooper 
1955). Progress in this area of physiological oppositions can be 
found in Peterson and Shoup 1966 and Perkell 1969. I return to this in 
greater detail in Chapter 8. 

From the standpoint of constructing a phonology from observable 
phonetic data, then, the notion of the opposition maintains a prominent 
position. Not only does the opposition command the necessary link 
with the phonetic data, but it also creates a convenient real unit 
for organization of the structure into a functional system. This is 
accomplished through the Prague notions of opposition theory. 

According to Prague opposition theory (as found in Trubetzkoy 1969 J 
Jakobson 1962; etc.), the various oppositions can be divided into two 
types (leaving the morphophonological abstractions aside). As I mention 
In section 6.3, these types are known as the phonology (those oppositions 



with phonological pertinence) and the phonetics (those oppositions with 
phonetic perinence only). In the phonology, the oppositions serve a 
contrastive function and predict the particular members of the phonetic 
oppositions. Insofar as the particular members of the phonetic opposi- 
tions are predictable from the contrastive phonological oppositions, 
the phonetic oppositions can at times serve (albeit indirectly) some 
contrastive function as well (compare Jakobson, Fant, and Halle 1952 i8). 
Hence, as long as the phonological oppositions do serve the contrastive 
function, the level of pertinence of a particular opposition in a 
particular language may depend more upon the principle of predictability 
than upon that of contrast. As always, though, such a determination can 
only be made from considerations of the structure of that particular 

The relationship between the oppositions of the phonology and those 
of the phonetics, moreover, is one of abstraction. As I point out 
above, the oppositions have their basis in observable acoustic and 
physiological phonetic data. The characteristics of these data which 
consistently and predictably occur are abstracted on the phonetic level. 
These characteristics become the oppositions of phonetic pertinence. 
The oppositions which consistently and predictably occur with contrastive 
function and which can be used in the prediction of other opposition 
members are abstracted on the phonological level. Those oppositions 
which then consistenly and predictably cooccur with elements of grammatical 
function are abstracted on the morphophonological level. 

The process of abstraction outlined above is consistent with 
that mentioned in section 6.3. It maintains the Inner approach to 
phonology, defines its relationships in a manner at least not incon- 


slstent Kith the phonetic data, and abstracts such that whatever is 
abstracted on a higher level (that is, a more abstract level) is present 
on the lower (less abstract) level of abstraction. Thus, a minimum of 
assumptions is introduced to the phonology* 

We have been concerned throughout this work with the consonant 
system of Welsh, and particularly with the relationships that obtain 
between the stop cognates. As Trubetzkoy (l969i?6-7) points out, there 
are two oppositions that have a special relationship in stop cognates 
in many languages. These are the oppositions of voice and tension. In 
the stops of these languages, one of these two oppositions is of phono- 
logical pertinence and the other of phonetic pertinence only. In 
positions of neutralization, where one of the members of the phono- 
logical opposition is realized, the weaker, unmarked member occurs. 

For example, in German we find a position of neutralization for 
obstruents in word-final position. Thus, the /t/ in Land /lant/ 
•land (nominative) 1 alternates with the /d/ in Lande /lande/ 'land 
(dative)', such that the final stop is voiceless and (in Standard German) 
fortis. In Welsh, on the other hand, we find a position of neutraliza- 
tion for stops in word-final position which is realized in precisely 
the opposite manner from the German. The /g/ in teg /teg/ 'fair' 
alternates with the /k/ in teced /teked/ 'fair (equative)', such that 
the final stop is voiced and lenis. Whereas the voiceless member of 
the opposition of voice is the unmarked and the lenis member of the 
opposition of tension is the unmarked, we can conclude that voice is 
the phonologically pertinent opposition in German and tension is the 
phonologically pertinent opposition in Welsh (compare the »principium 
divisionis' of Jakobson 1962 t23). 


As mentioned in section 6,2, sound change, as well as neutraliza- 
tion, occurs with respect to the unmarked member of the phonologically 
pertinent opposition. Once we recognize that Welsh is a 'tension 
language*, it is not difficult to accept a change in the system in which 
the voiced stop replaces the voiceless, for voice in such a system is 
of phonetic pertinence only. 

The notion that the basic opposition in Welsh stops should be 
tension, with fortis and lenis members, ought not to cause much surprise 
among Cymrists, In fact, this is the traditional interpretation of the 
consonant system. A change from the voiceless aspirated stop to the 
voiced unaspirated stop is known historically and in the mutation 
system as lenition (compare the term treiglad meddal 'soft mutation' ), 
while a change in the opposite direction is known as provection 
( calediad 'hardening'). 

In spite of its convenience and its traditional background, the 
opposition of tension and the notions of fortis and lenis are difficult 
to define phonetically, especially that type of tension often described 
as being pertinent to consonants. Tension is a cover-term for an 
opposition which is realized in different ways in different languages. 
Just how it is realized in Welsh demands an examination of the language. 

At this point, I should like to return to the manner of neutraliza- 
tion in German as opposed to that in Welsh. In German, the marked 
phonological opposition member of the opposition of voice is realized 
as the voiceless unmarked member; while the phonetic opposition member 
of the opposition of tension, which would be unmarked if German were 
a tension language, is realized as the fortis member, which would be 
marked if it were a tension language. On the other hand, in Welsh, 


the marked fort is phonological opposition member is realized as the 
unmarked lenis member; while the phonetic voiceless opposition member, 
which would be unmarked if Welsh were a voicing language, is realized 
as the voiced member, which would be marked if it were a voicing language. 
Thus, the oppositions of voice and tension in Welsh, as well as in 
German, are complementary. 

As Welsh has only two phonologically pertinent classes of stops, 
we can consider (for the moment) the opposition to be privative, capable 
of being described with binary features (presence versus absence of 
singulary features). If we turn, then, to the specifications of features 
in Table 1.0. c, we find that one feature is indeed complementary to 
voice. This feature is aspiration. Because the opposition described 
as tension is complementary to voice and aspiration is complementary to 
voice, we might consider that tension is realized in Welsh through the 
opposition of aspiration. Of course, a conclusion that tension includes 
or consists of aspiration cannot be made on the basis of these facts 
alone. Nevertheless,, in this chapter I demonstrate that aspiration is 
indeed the opposition through which this cover-term of tension is 
realized in Welsh, and that aspiration is the basic opposition not only 
of the stops, but also of the entire class of consonants. 

In order to determine the place of aspiration in the Welsh con- 
sonant system (or subsystem), it is necessary to determine first just 
what aspiration is. This entails an investigation of the nature of 
aspiration in the phonetic system of Welsh and the relations which 
hold between aspiration and the other oppositions in the phonology 


7.1 The Prosodlc Nature of Aspiration . It Is mentioned in the previous 
section that tension is a difficult opposition to define, as various 
languages realize tension in various ways. The opposition of aspiration 
is also difficult to define even within one particular language. It 
involves apparently a release of pressure somewhere in the vocal 
apparatus, but Just where this pressure is exerted and how it is released 
is open to debate. In Chapter 10, I treat this problem in some detail 
from an acoustic and a physiological standpoint. 

Wherever this aspiration emanates from, it is realized in a manner 
similar to the glottal fricative /h/. It can be realized alone in 
syllable-initial position, in which case it is known as preaspiration 
(before the nuclear vowel). On the other hand, it may be realized in 
conjunction with some other consonant, in which case it is considered 
to be the aspiration of that consonant. 

When we speak of a particular opposition member as being realized 
in a particular consonant (for example, an aspirated stop as a stop with 
aspiration), we mean that that opposition member is inherent in the 
consonant. An inherent opposition is just one of two basic types of 
opposition. The other type is the prosodic opposition. Whereas the 
inherent opposition is thought to belong to a particular segment (or 
phoneme), the prosodic opposition occurs over segmental boundaries. To 
be sure, the range of a prosodic opposition depends to a great extend 
upon the segments with which it interrelates in the syllable, but, 
unlike the inherent opposition, it is not bound to the range of the 
segment. The nature of a prosodic opposition is most easily seen In 
stress, pitch, and tone (in fact, these are the only prosodies given 
by Trubetzkoy 1 969) --the suprasegmentals or secondary phonemes of 


Bloomfield (1933 »90). 

The nature of the opposition of aspiration, whether it is inherent 
or prosodic, can most clearly be seen in some phenomena involving 
voicing and aspiration in word-initial consonant clusters. As can be 
seen in the data of Fynes-Clinton 1913i wnen a liquid follows a word- 
initial voiced unaspirated stop, it is voiced. Immediately, the notion 
of noncontrastive distribution comes to mind. But it is nonetheless 
conceivable that it is merely coincidence that voiceless liquids follow 
voiceless aspirated stops while voiced liquids follow voiced unaspirated 

The mutation system, as presented in Chapter 1, shows that this 
relationship is far more than just coincidence. As we see in soft 
mutation rule 1.1. a, in environment II a voiceless aspirated stop is 
realized as a voiced unaspirated stop. The voiceless liquid following 
the radical, moreover, is realized in soft mutation as the voiced 
cognate. For example, we find the following forms in Fynes-Clinton 
1913 (entered in alphabetical order) i* 














soft mutation 



party (political) 
























soft mutation 

grit st 



An initial stop can also be followed by a nasal /n/. When this 
nasal follows the voiceless stop, however, it is voiced, as in the 
following data (also Fynes-Clinton 1913) » 











When the spirant mutation rule 1.3 applies to a voiceless aspirated 
stop preceding a voiceless liquid, the radical is realized as a voice- 
less fricative, but the liquid is realized as the Voiced cognate. Thus, 
we find the following (Fynes-Clinton 1913) « 



















crowdy (fiddle) 

At this point, we can make a generalization. The presence of 
voice in the liquid is determined by the presence of voice (and aspira- 
tion) in the stop, but it is not dependent upon the absence of voice 
in the initial fricative. 


In the nasal mutation rule 1.2. a, we find a more complex situation. 
The voiceless aspirated stop is realized as a voiceless nasal with the 
aspiration heightened to the degree of a glottal fricative /h/, as 
though this glottal fricative were a separate segment as in preaspiration. 
When the consonant realized as the nasal mutation form of a voiceless 
aspirated stop is followed by a liquid or by a nasal, the liquid or 
nasal is not only voiceless, but it is followed by the aspiration of 
the initial segment— the glottal fricative. Thus, we find the following 
(Fynes-Glinton 1913) » 


pie id 




klis st 





nasal mutation 




nihil st 







dead of night 



diminutive epithet 

If we are only concerned here with inherent oppositions, then we 
should be able to handle these and like phenomena through a segmental 
description, such as that found in generative or structural phonology 
(phonemics). Let us first, then, treat the data from the assumption 
that whatever alternations occur do so through the implications of 
inherent oppositions only. 

Considering first the soft mutation data, it would appear as though 
we have nothing more than a case of voice assimilation. The voice of 
a liquid is determined by the voice of the preceding stop. In generative 
terms, this could be handled by a rule, making use of the •alpha-switching' 


notational device (see Griff en 1975)t Such a notion is Insufficient, 
for we also find such items as the following (Fynes -Clinton 1913) » 









In these items, we find a voiceless (unaspirated) stop preceding a 
voiced liquid. In fact, there are two voiceless consonants Before the 
voiced liquid. 

From the above items, it would appear that aspiration is involved. 
The /s/-plus-stop environment is a well-known deaspirating environment 
in English (see Gleason 196l«263» Kurath 1964 i?^) as well as in Welsh 
(see Fynes-Clinton 1913ixxii). From these data we might consider that 
the aspiration of the initial consonant determines the voice of the 
following sonorant. Such a notion as the opposition of one segment 
determining a different opposition of another, however, is actually 
prosodic, for it requires an opposition to dominate a range irrespective 
of segmental boundaries. Thus, in a segmental solution relying only 
upon inherent oppositions, we cannot determine the voice of one segment 
by the aspiration of another — we must search out another solution. 

We could modify our voice assimilation by further restricting 
the environment of its application. We could specify that it only applies 
where the stop preceding the liquid is in word-initial or syllable- 
Initial position. In order to avoid other deaspirating environments, 
we might find it necessary to further restrict the environment of 
application through stress. 


When we consider the phenomena surrounding nasal mutation, however, 

the segmental, inherent-opposition solution falters. There are two 

problems in accounting for the alternation of, for example, [pnelin] 

and [mnhelin] through inherent oppositions only. Nasal mutation will 

produce an initial cluster /mhn-/. The problems are (l) how to have 
the aspiration realized on the other side of the /n/ and (2) how to 
account for the devo icing of the /n/. 

In solving the first problem through inherent oppositions, we 
would have to rely on metathesis. Once an opposition is removed by 
two segmental boundaries from the segment to which it belongs, though, 
we are not really operating with inherent oppositions. Furthermore, 
if we cannot use prosodic oppositions, accounting for the voiceless 
/n/ becomes a formidable task. As we see in the stop -plus -nasal clusters, 
above, the voicelessness of the initial consonant is not sufficient to 
cause the nasal to be realized as voiceless. Yet, in this case it must 
be sufficient. These two problems, especially when taken together, are 
so complex as to render the solution with inherent oppositions inoperable. 

On the other hand, by making use of aspiration as a cluster prosody 
(compare Robins 1957), we can account for these phenomena quite simply. 
Given the notion that, by definition, that which is aspirated is voice- 
less (at least, in Welsh), we can account for the data by adopting the 
notation of prosodic analysis (see Firth 19^8), as follows i 

h h h_ 


This is to say that aspiration from an aspirated (voiceless) stop 
dominates the following liquid but not a following nasal} while that of 
an aspirated nasal dominates the following sonorant and the onset of 


the vowel. Where an /s/ precedes the cluster, the aspiration would 
emanate not from the stop, but from the /s/, allowing us to maintain 
a voiceless but unaspirated stop and a voiced liquid. 

In our Prague-oriented method of abstraction into a level of 
phonological pertinence, outlined in the previous section, this develop- 
ment is particularly important. The opposition of voice can always be 
predicted from the opposition of aspiration, but not vice versa. This 
shows that of the two, aspiration and voice, though both could serve 
a contrastive function (at least indirectly), aspiration satisfies the 
predictability requirement. I return to this notion in the next section. 

In an unpublished manuscript, Lockwood (ms.) incorporates the 
Firthian notion of prosodic analysis into the stratif lcational model by 
means of the 'quasi-step matrix' (compare the matrix of Lamb 1966b). 
This device reflects the prosodic nature of the dominating opposition 
(or component) in the hypophonemic stratum by allowing components to 
overlap several columns in the matrix, thus reflecting the phonetic 
fact that speech is continuous rather than the utterance of one group 
of components after another. 

In Figure 7.1, we see that this innovation in the matrix of 
/mnhelln/ not only incorporates the Firthian notion of prosodic domination, 
but it also incorporates the Prague notion of contextual neutralization 
and assimilation of oppositions (see Trubetzkoy 1969). In Griff en 1975, 
I demonstrate how generative phonology can also describe the same 
phenomena through using aspiration as a prosodic opposition. No matter 
what framework is used, however, only by recognizing the prosodic nature 
of aspiration in Welsh can the data be adequately represented. 



















Figure 7.1 
Quasi-Step Matrix of /mnhelin/ 


The implications of the basic opposition of a language being 
prosodic rather than inherent are quite far-reaching. As we see in 
the next section, we can expect relations between various segments in 
the phonological system to apply without respect to the boundaries of 
these segments. For example, as aspiration associated with a par- 
ticular stop is normally explosive (that is, being realized after closure), 
an aspirated segment following a stop is liable to have considerable 
effect on that stop. 

More than this, however, the notion that the basic opposition of 
Welsh should be prosodic and should operate without regard for segmental 
boundaries calls to question the very notion of segments themselves, 
as meaningful units of phonological structure. Certainly, the fact 
that the segmental boundaries can be disregarded by the basic opposition 
raises the question as to whether all boundaries can be disregarded by 
any opposition. In other words, if the basis of the consonant system is 
nonsegmental , what of the consonant system itself? I return to this 
question in section 7.3 and in Chapter 8. 


7.2 Aspiration in the Phonological System . In the previous section, 
we find that aspiration determines voice in sonorants In certain con- 
sonant clusters. The aspiration of any aspirate consonant is sufficient 
to devoice a following liquid in a cluster, and heightened aspiration is 
sufficient to devoice a following nasal (or, more precisely, to account 
for the realization of the voiceless cognate in such clusters). Such 
evidence is necessary to show that tension, which determines voice, is 
realized through aspiration, but it is not sufficient to show that the 
opposition of aspiration is the basis of the consonant subsystem of 
this tension language. After all, similar phenomena can be found in 
German, a voicing language (compare section 6.2), 

In order to demonstrate that (at least for all practical purposes) 
the opposition of aspiration is the medium through which the tension 
relationships in the phonology are realized, and that, as such, it forms 
the basis of the phonological relationships in the consonant subsystem, 
it is necessary to examine the structure of the consonant subsystem 
within the phonology and the manner in which this structure interacts 
with the opposition of aspiration. As I point out in section 6.1, the 
structure of those consonants which enter into any relationships with 
other consonants (in the Standard Welsh, this includes all consonants 
except /s/3) is organized along the lines of the fortis-lenis scale, 
as found in Table 6.1. As Welsh is a tension language (as shown through 
its markedness relationships in section 6.2), an organization of the 
structure on the fortis-lenis scale is expected. 

A more complete structure of the consonant subsystem can be found 
in Table 7. 2. a. This structure takes into account the nasals, which 
are related with the structure in such a way that columns 2 and 3 may 


be considered as parallel to the stops in columns 2 and 3. For ease 
of conceptualization, we might consider the nasals of columns 2 and 3 
positioned above the stops in a three-dimensional structure, such that 
a stop in column 2 will relate not only with the consonants in columns 
1 and 3, but also with the nasal in column 2. To the basic relation- 
ships of the opposition of tension, this adds those of the opposition 
of nasality (compare the superimposed bundles of Trubetzkoy 1969 »88). 

Table 7. 2. a . 
The Consonant Subsystem of Welsh 

12 3^ i2 

v b p f m mh 

r o 

6 d t 8 n nh 


g k x o gh 

1 1 

As I mention in section 6,1, such a scale bears a strong resem- 
blance to the strength scales of Foley 1970a. In fact, it is identical 
to his beta-strength scale in its fortis-lenis relationships. Although 
the strength scales of Foley 1970a are totally useless from the stand- 
point of determining the basis of relationships, they are nonetheless 
quite useful in that they do summarize important synchronic alternations 
and diachronic changes on the phonological (or at least graphemic) 

The synchronic phonological relationships between the columns are 
the following! Column 2 and column 1 are related through soft mutation 
(section 1.1) i column 3 and column 2 are related through soft mutation j 
column 3 and column b are related through spirant mutation (section 1.3) J 


the two columns 2 and the two columns 3 are related through nasal 
mutation (section 1,2), In this section, I demonstrate that the 
horizontal relationships in Table 7. 2. a can be explained through the 
opposition of aspiration. 

Relationships between elements in the gradual opposition of tension 
(the fortis-lenis scale) are categorized into two types. As is the case 
with the mutations, these relationships are traditionally couched in 
terms of processes — the process of lenitlon and the process of provec- 
tion. Traditionally, lenitlon consists of a change from a stronger to 
a weaker opposition member (from one number in Table 7. 2, a to a lower 
number), while provection involves a change from a weaker opposition 
member to a stronger one (from a lower number to a higher). 

The first element of lenition to be considered is the notion of 
soft mutation. In traditional terms, the mutation of a voiceless aspirated 
stop radical to its corresponding soft mutation form involves a mutation 
to a more lenis consonant. Now there are two oppositions involved — voice 
and aspiration. As a change from voiceless to voiced in Welsh would be, 
if anything, a strengthening along the fortis-lenis scale, from the 
observable oppositions in the change, it would appear as though deaspira- 
tion is involved. In this case, deaspiration is at least isomorphic 
with lenition. 

An intersting development in the history of soft mutation is the 
creation of the consonants 11 /l/ and rh /*/ (in a manner similar to 
the development of the affricate subsystem— see Griff en 197^b), Due to 
their behavior in borrowings as lenis initials (see Griff en In press a), 
1 /l/ and r /r/ can be seen to be members of column 1, According to 
Jackson (l953»^73-80), where Brythonic maintained a non-lenited liquid 


(a liquid not in the environment of lenition — at that time the environ- 
ment was phonological), by around the tenth century, Welsh developed the 
lateral and trill fricatives. In Modern Welsh, this process continues 
in the corresponding grammatical environment, such that a word borrowed 
with a column-1 initial consonant will eventually change such that the 
initial consonant is realized as its column-2 cognate except, of course, 
in soft mutation. 

The relationship between the liquids and their corresponding 
fricatives can easily be accounted for through aspiration. As in the 
relationships between the column-3 and column-2 stops, voicing is not 
primary, as this process would run counter to lenition. The question 
then becomes, 'What is the difference between /l/ and /l/, and /»/ 
and A"/?' Taking the change in voice into consideration, a strong 
case for analogy could be made between the lenition of voiceless aspirated 
stops and that of voiceless lateral and trill fricatives, establishing 
the pertinent difference to be one of aspiration. As aspiration is some 
sort of release of pressure and this release would be realized through 
a glottal -fricative-like sound, a relationship between a liquid and a 
fricative variant can readily be seen through pressure added to the 
continuance of the liquid. The orthography gives us another indication 
as to the basis of the relationship being aspiration, as the forms have 
often been written (up to the standardization of the alphabet in the 
early twentieth century) as lh and rh. Indeed, the trill fricative is 
still written today with the h. It is not, however, through lenition 
that the relationship becomes most clear, but through provection, as 
I demonstrate below. 

The relationship between the column-3 consonants and the column-2 


consonants Is based upon aspiration, as Is the relationship between the 
lateral and trill in column 2 and those in column 1. This gives us a 
basis for two analogies to establish aspiration as the basis of the 
relationship between the remaining consonants in columns 1 and 2. 
Unfortunately, the strength of such an argument relies wholy upon the 
strength of analogy as a method of extension. Such evidence, then, is 
hardly conclusive ; although, to be sure, it is evidence. 

Another analogy can be made from the relationship between the 
consonants in columns 2 and 3 in the position of neutralization. In 
Middle Welsh, the letters p_, t, and c (or k) were used in word-final 
position to represent the voiced unaspirated stops in the position of 
neutralization (see section 5.1). In the same position, the letter d 
represented the voiced dental fricative /&/ (see Evans 1960«5t Watkins 

1961 i78). 

Now at this point, we ought to consider what causes a stop to be 
realized as the voiced cognate in word-final position. According to 
Pilch 1958, even where a voiceless stop does occur in word-final 
position, it is not aspirated. For example, Pilch (I958t40) transcribes 
twp 'stupid' as /2'thup/, showing that aspiration may occur in initial 
position, but not in final position. 5 The reason why word-final position 
should be a position of neutralization of the opposition of aspiration 
would appear to be the relative difficulty of pronouncing an aspirated 
sound at the end of a word in continuous speech (compare Trubetzkoy 
1969i262, in which the correlation member most frequently found in 
continuous speech should be the unmarked). 

This neutralization of aspiration in word-final position Indicates 
two things. Firtt, the realization of an archiphoneme on the basis of 


the markedness conditions in the opposition of aspiration should indicate 
that it is the opposition of aspiration (or an opposition isomorphic 
with it) which serves as the phonologically pertinent opposition between 
the two classes. Second, if the position of neutralization is dependent 
upon the opposition of aspiration in this case, it may also be dependent 
upon the same opposition in the same environment in the case of the 
realization of d as /6/. Again, because of the nature of analogy, 
the latter evidence is necessarily weaker than the former. 

The relationship between aspiration and the various degrees of 
the fortis-lenis scale is best demonstrated through the process (tradi- 
tionally speaking) of provection. There are many different types of 
provection in Welsh. The one type, though, which shows the closest 
connection with aspiration is one which I call 'aspirate provection'. 

In Modern Welsh, aspirate provection often surfaces in a type of 
poetry known as cynghanedd . The complex rules of cynghanedd require 
patterns of alliteration between the first half of a line of verse and 

the second. For example. A. LI. Roberts 1973 «15 cites the following 


line of verse r 

Ochain cloch i a chanu clir 



In the aspirate provection, when a voiced unaspirated stop precedes 
an /h/, the stop becomes the voiceless aspirated cognate. A. LI, 
Roberts 1973 1*7 cites the following lines « 

Y poen i yn ei hwyn eb hi 



Onid hen i yw cerddi'n tud? 

- 1 r 

A thrig hedd i uwch llethrau cwm 

The /h/ represents aspiration in its simplest form. When this aspira- 
tion precedes a stop, there is a shift in syllable boundary so that 
the glottal fricative becomes the explosive element of the stop. Voice, 
being the phonetically pertinent opposition, is determined by the 

More important than the provection caused by the following /h/ is 
that caused by a following rh /*/. A. LI. Roberts (l973»^7-8) also cites 
the following correspondences in the cynghanedd l 

Y mab rhad t mwyach a'm pryn 

Onid rhyw awr t yw'n byw trist? 


Mae eisiau cred i ymysg rhai 

cred i ymysg rna 

The relationships illustrated in these correspondences not only estab- 
lish aspiration as the opposition linking the consonants of column 2 
with those of column 3, but they also demonstrate the aspirate nature 
of the fricative cognates of the liquids. Once the aspiration is 
removed from the fricative in order to 'harden' the voiced unasplrated 
stop, the remaining oppositions of /*/ are identical to those of /r/. 
Thus, the relationship that obtains between at least half of the con- 
sonants of column 1 and their cognates in column 2 can be seen to be 
nothing else but aspiration. This adds considerable weight to the 


analogy that would link the remaining two consonants of column 1 to 
their column-2 cognates through aspiration, as well as, indirectly, 
it supports the other two analogies, 

A particularly interesting application of aspirate provection is 
found in the creation of family names. In the Middle Ages, Welshmen 
did not generally use family names. A man would use the familiar 'son of 
formula, as in the name of the famous knight Owen ap Urien (Owen son of 
Urien). This aj> 'son of is pronounced /ab/, due to the nature of the 
word-final position of neutralization. At the end of the Middle Ages, 
when (in accordance with decrees of the English occupation) family names 
were taken, many Welshmen simply took the final consonant of ap_ /ab/ 
and added it to the initial vowel of their fathers' names. Thus, a 
man who had been known as, for example, Idris ap Owen (idris son of 
Owen) might have taken the name Idris Bowen. Where a father's name 
began with an /h/, however, we find aspirate provection, such that, 
for example, Idris ap Hugh might have taken the name Idris Pugh, and 
Idris ap Howell might have taken the name Idris Powell. Moreover, 
where the father's name began with /*/» w * find t* 1 ® &*&* aspirate 
provection as that in the poetry, such that, for example, Idris ap 
Rhichard (or Rhisiard) might have taken the name Idris Prichard (or 
Prisiard), and Idris ap Rhys might have taken the name Idris Prys 
(or Price), 

This aspirate provection is not limited to poetry and family 
names. It is quite productive in the spoken language, and Fynes-Glinton 
(I913txviii) notes that it is found in the Bangor district dialect. 
For example, we find the following pronunciations t 





ei mab hi 

i rahat pi 

her son 

ei thad hi 

i 9a: ti 

her father 

ei cheg hi 

i xei ki 

her mouth 

We should take special note of what might appear to be a 'shift' in 
syllable boundaries, for it shows a realization of a voiced unaspirated 
stop along with a unit of aspiration. The only way that such a phenomenon 
could occur is through the prosodic nature of the opposition of aspira- 
tion, a nature which allows a segment (/h/) to effectively merge with 
another segment without regard to the supposed segmental boundaries. 

This nature of aspiration as a prosody, demonstrated in the previous 
section, becomes extremely important with another type of provection. 
Not only does an occurrence of aspiration following a voiced unaspirated 
stop result in the realization of a single voiceless aspirated stop, 
but an occurrence of two identical voiced unaspirated stops in one 
position in the syllable (onset or coda) also results in the realization 
of a single voiceless aspirated stop. This type of provection I call 
•geminate provection'. 

The geminate provection has been productive since the transition 
period between Brythonic and Old Welsh, where it culminated in the loss 
of syllables, For example, Morris Jones (1913 »182) cites the Brythonic 
name Gato-tigirn- /katotigirn/ as developing into * Cad-diYirn - 
♦/kaddiYirn/, than (^tteyrn /kat(t)eirn/, and finally Cateyra /kateirn/. 
In the Modern Welsh dialect of the Bangor district, Pynes-Cllnton 
(I913ixxiil) cites the following j 


orthography transcription gloss 

pob blwyddyn poip plu»46in every year 

gwybod dim gubot tlm not to know 

tebyg gennyf tebik kin 1 I suppose 

Whether there Is a shift In syllable boundary here Is not given by 
Fynes-Clinton. In the case mentioned above, the shift in boundary is 
mentioned in passing, only after all forms are given. Although no 
mention is made here, a shift is probable, taking into consideration 
the nature of provection. 

Geminate provection can also arise when a prefix ending in a voiced 
unaspirated stop precedes a root beginning with an identical stop. 
For example, the prefix lied- /led/ 'half may be affixed to the noun 
tyb /tib/ 'opinion'. After the application of soft mutation (see Evans 
and Thomas 1968i452), the result is lletyb 'leUb/ 'doubt', which may 
alternatively be realized in the verb-noun form as lled-dyblo /ledctobyo/ 
(compare Bwrdd Gwybodau Celtaidd 19^2 »95). 

In order to understand this type of provection, it is necessary 
to consider the nature of aspiration as a prosody. Working within a 
framework which recognizes the segment, it is not easy to visualize 
the relationships involved, and I return to this notion in Chapter 9. 
Nonetheless, we can see that the development of this phenomenon involves 
placing two identical stops into one slot in the syllable. If the stops 
were not homorganic, we could get a cluster | but so long as they are 
homorganic, they cannot be pronounced as geminates in a single position 
in a syllable (see section 8.2)— the nature of stops as elements of 
closure does not allow this. 

Although a stop involves a single act of closure, it may also be 


accompanied "by a prosody. Now if the prosody constitutes a gradual 
opposition, then the prosody is realized in each member of the opposition 
in a scale of gradual presence. From Table 7. 2, a, we can see that the 
phonological relationships in the consonant subsystem are organized 
along the lines of a gradual opposition. For the realization of two 
stops from column 2 sis one stop from column 3» we would need some 
opposition present in both of the column -2 stops to combine so as to 
create the level of strength found in column 3. Of course, in order 
for such an opposition to combine without regard to boundaries, the 
opposition would have to be prosodic. 

Thus, there is some sort of prosodic opposition which parallels 
the system shown in Table 7. 2. a, at least as far as columns 2 and 3 are 
involved. It is difficult, however, to say that this opposition is 
aspiration, as we would assume that some degree of aspiration is present 
in the column-2 stops, and aspiration (as it is usually conceived in 
the sound of [h"J) is not readily perceived in these voiced unaspirated 
stops. This does not mean, however, that aspiration could not be 
realized in different perceptual ways in different degrees of the 
opposition, I examine this phonetic problem in Chapter 10. 

The relationship between this type of provectlon and the aspirate 
provection, which is definitely a case of the opposition of aspiration, 
is brought out in the poetry of the cynghanedd . The geminate provection 
of two voiced unaspirated stops is not normally used in alliteration 
with the voiceless aspirated stop but with the realization of the 
aspirate provection. A, LI. Roberts (1973 «^8) cites the following 
lines of verse « 

Byd hebhedd i nid byd hebboen 

Onid hardd t dy henwlad di? 

Deg rhodd j y godid og Gr ist 

This practice of the cynghanedd shows that the relationships 
that obtain between the voiced unaspirated stops of geminate provection 
and the voiced unaspirated stop and aspiration of aspirate provection 
are either the same or isomorphic. The prosody involved in geminate 
provection, then, is either the opposition of aspiration (which is the 
only prosody that has been (or can be) demonstrated from the data 
associated with the fortis-lenis scale) or some opposition isomorphic 
with it. By 'isomorphic' I mean 'sharing all pertinent characteristics', 
indluding the nature of aspiration in the system of relationships that 
hold in aspirate provection (the [hj-sound possibly being phonetic, not 

It is important to appreciate the (at least) isomorphism of the 
relationships that obtain in one case in the voiceless aspirated stop, 
in another case in the combination of a voiced unaspirated stop and 
aspiration, and in yet another case in the combination of prosodies 
through the realization of two identical voiced unaspirated stops as 
a single voiceless aspirated stop. Comparing the last-mentioned rela- 
tionship with the first two, we find the aspect of the gradual opposition 
that is most pertinent to the aspiration hypothesis. This aspect 
consists of two points t (l) The gradual prosodic opposition that 
forms the basis of the consonant subsystem behaves in such a manner that 
the combination of two units of prosody (that is, two levels of strength) 
in a realization of a consonant in column 2 results in the realization 


of the cognate consonant from column 3* and (2) the nature of this 
prosodic opposition is identical to the nature of aspiration. 

These two aspects of the gradual opposition forming the basis of 
the scale of relationships found in Table 7. 2. a gain particular meaning 
when viewed in conjunction with the development of the voiceless frica- 
tives, the elements of column k. According to Jackson (1953 « 565-73 )t 
the voiceless fricative developed during the Brythonic period through 
the same process that we find in the geminate provection. When two 
identical voiceless stops from column 3 combined, the result was a 
single realization of the cognate from column k. 

Again, if two identical stops come together in one slot in the 
syllable, the nature of the closures associated with the stops cannot 
be realized in a cluster. The prosodies, however, can combine. In 
the voiceless aspirated stop, moreover, the prosody can be perceived as 
the [h]-sound of aspiration. Thus, the member of the prosodic opposi- 
tion that combines in geminate provection from column 2 is realized in 
the combination as the aspiration i while this aspirate member of the 
gradual prosodic opposition that combines in geminate provection from 
column 3 is realized through voiceless fricatlon, which may be perceived 
as an exaggeration of the [h] -sound. 

This notion is supported in the development of the othography. 
At least in the case of /©/ and /x/, the spelling of these consonants 
was rendered (sporadically, to be sure) by the letter for the voiceless 
aspirated stop followed by h as early as the late sixth or early seventh 
century. For example, we find the Brythonic inscription MACGODECHETI 
from that period (Jackson 1953 »566). Indeed, even after the spelling 
reforms of the early twentieth century, the spirant mutations derived 


from the voiceless aspirated stop radicals are still rendered in this 
manner, even in the case of pji /f/, representing a sound written as 
ff /f/ elsewhere. 

Examples of geminate provection of voiceless aspirated stops can 
be found in words derived from Latin and Brythonlc. Welsh cyff /k*f/ 
•stem' is derived from Latin clppus /kippus/, Welsh Brython /breQon/ 
'Briton' is derived from Brythonic Brlttones /hrittones/, and Welsh 
pechod /pexod/ 'sin' is derived from Latin peccgtum /pekkaitum/ 
(Morris Jones 1913 tl75). 

Further evidence of the presence of aspiration in the consonant 
subsystem can be found in the aspirate mutation. As we see in section 
1.4, this mutation involves the addition of preaspiration to an initial 
vowel. In Standard Welsh, this mutation only affects vowels, but in 
some dialects, including the dialect of the Bangor district, the glides 
and the labial nasal are also affected. In Fynes-Clinton (l913«xvili), 
for example, we find the following! 7 




el (h)iaith 

i hyaie 

her language 

el (h)wats 

i hwac 

her watch 

ei mab hi 

1 mhat pi 


her son 

In the last item of the above examples, a consonant of the nasal 
column 3 is found as a result of the addition of aspiration in aspirate 
mutation to its cognate from nasal column 2. The relationships that 
hold between the nasal columns of Table 7. 2. a can thus be seen to 
consist wholy of aspiration. 

There is one final piece of evidence to begleaned from the aspirate 


mutation. We should take note that one of the most common environments 
of aspirate mutation, that mentioned in the preceding examples, is 
identical to one of the most common environments of spirant mutation, 
environment IVb (compare also section 2,8), This, combined with the 
nature of the development of the voiceless fricative in Brythonic, 
may give added credence to the notion that aspiration still forms the 
basis of the relationship between the consonants of column 3 &nd those 
of column 4 in Modern Welsh. 

One final case of provection can be found in the behavior of 
lenis initials in Welsh borrowings (see Griff en In press a). As I 
mention above, when a word is borrowed into Welsh with an initial con- 
sonant from column 1 of the Table 7, 2. a, this lenis initial •hardens' 
to the cognate from column 2, The reason for this is that word initial 
position is a position of relative strength, insofar as the relation- 
ships between the consonants of the gradual opposition are concerned. 
Due to this position, the provection occurs. For example, English 
velvet has been borrowed as melfed , Latin liber /liber/ •book' has been 
borrowed as llyfr /livr/, and English rack has been borrowed as rhac 
/*ak/ (or [sag]), etc. As in the case of lenition from the column-2 
consonant to the column-1 consonant, the influence of aspiration in 
cases other than the liquids can only be ascertained through analogy. 

Through the processes of lenition, we can account ferine relation- 
ships that hold between the consonants of column 2 and those of column 
1, as well as for those relationships that hold between the consonants 
of column 3 and those of column 2. Through the processes of provection, 
we can further account for the relationships that hold between the 
consonants of column 1 and those of column 2, between the consonants 


of column 2 and those of column 3 ( as "ell as between the nasal columns), 
and between the consonants of column 3 and those of column 4. 

Furthermore, for all relationships between the various members of 
the gradual opposition except two of the four consonants of the lenis- 
most opposition member, there is direct evidence to support the notion 
that the prosody underlying the relationships is aspiration. In those 
two exceptions, there are several analogies to give indirect evidence 
to support the same notion. 

Thus, it would appear that the gradual opposition through which 
the consonant subsystem of Welsh phonology is structured is either the 
prosodic gradual opposition of aspiration or some opposition Isomorphic 
with it. By making use of the notion of phonological pertinence versus 
phonetic pertinence, the basic foundation of our abstraction system, 
we can effectively merge these two choices into one. This can be done 
by maintaining that the [h]-sound of aspiration is a phonetic character- 
istic of the phonological opposition of aspiration realized in certain 
conditions. Such a stance is consistent with the inner approach, for 
in reality, all sounds are phonetic In nature. Just as the vibrations 
of the vocal cords in the realization of certain members of the opposi- 
tion are phonetic but not phonologically pertinent, there is no reason 
why the [h]-sound could not be a phonetic symptom of aspiration but not 
the phonologically pertinent basis. In Chapter 10, I investigate the 
question of just what the common phonetic elements of this gradual 
opposition are. 

Taking this phonological opposition of aspiration to be the basis 
of the consonant subsystem of Welsh, I reconstruct Table 7. 2. a into 
Table 7.2.b. This new table brings out several points important to this 


analysis. First of all, it Is structured along an order-and-series 
framework (compare Martinet 1952), in which the various orders can be 
seen to be functionally contrast ive positions of articulation. Phono- 
logically, the members of the opposition in each order vary only in 
the degree of phonological aspirate prosody associated with them. 
Because word- initial position has played such an important role through- 
out the past six chapters, column 2 of Table 7. 2. a is taken to be the 
zero (or basic) position in the scale, as this is the position to which 
the stronger members change in the alternations and developments of 
Chapters 3, 4, and 5 (whether or not these constitute enough evidence 
for a shift) and as this is the position to which the weaker members 
change in borrowings. No mathematical relationships are implied in the 
numbers used. 

Table 7.2.b 
The Consonant Subsystem 

of Welsh 



























In the notation of Table 7.2.b, each degree of aspirate prosody 
within the gradual opposition is represented by the symbol /h/, while 
the degree weaker than the basic level is represented by /ty. Nasality 
is represented by the symbol /n/. There are certain conventions con- 
nected with this system. For example, the symbol /gV is a place-filler 


for zero, and the subtraction of one degree of aspiration from /bn/ 
causes the /n/ not to be realized. 

This system of orders and prosodies (taking nasality to be prosodic — 
compare E.W. Roberts 1972) accounts for all relationships (including 
mutations) in the Welsh consonant subsystem with a maximum of uniformity. 
All lenltion is simply the reduction of one degree of aspirate prosody 
within a single order, provection is the addition of one degree, and 
nasalization is the addition of one degree of nasal prosody within a 
single order. All alternations and changes discussed in Chapters 1 
through 5 can thus be accounted for through the shift in one, and only 
one, prosody. One point must be stressed i Although I use symbols 
purely as a mneuraonlc labeling device to show a zero-degree and a 
minus -degree, the nature of aspiration as a phonological gradual 
opposition means that in reality the opposition is ever present in 
various degrees of realization. 

From the standpoint of the predictability principle, one point 
has become evident throughout this section. The gradual opposition of 
phonologically pertinent aspiration can always predict the phonetic 
occurrence (or optionality) of voice, but voice cannot predict aspiration. 
Moreover, in the neutralization of aspiration cited in Pilch 1958, the 
fact that it is aspiration which is lost (or, more precisely is not 
realized) further indicates that voice is a predictable phonetic opposi- 
tion. (Compare Watkins 196li24.) 

7.3 The Implications of a Prosodic Base . In the previous section, 

we see that the consonant relationships of Welsh can be handled through 

a structure organized in accordance with the opposition of aspiration. 


If this opposition of aspiration were an inherent opposition, there 
would be no particularly important implications in our using it. How- 
ever, as we see in the cluster phenomena in section 7,1 and in the 
realization of provection, aspiration is not an inherent opposition, 
but a prosodic one. 

The nature of the opposition of aspiration as a prosody makes 
this opposition particularly interesting in a phonological structure. 
The prosody operates without regard to boundaries of segments. Thus, 
we find the opposition, through the phonetic [h]-sound that charac- 
terizes it in particular opposition members, extending into segments. 
For example, the liquids of section 7.1 become devoiced by this aspira- 
tion following the voiceless aspirated stop that precedes them, and the 
nasals following aspirated nasals become devoiced and followed by the 
aspiration of the initial nasals. Likewise in aspirate provection, the 
following [h]-sound spreads its domination into the preceding stop, 
devoicing it. 

Furthermore, in the geminate provection, we find additional evidence 
of the disregard of this opposition for segmental boundaries posited in 
accordance with traditional notions. The prosodic opposition members 
in this type of provection coalesce as though no boundary existed. 

Moreover, in reality this prosody operates in the functional struc- 
ture outlined in Table 7.2.b as an independent entity. We could replace 
all positions of articulation with a variable x without upsetting the 
basic structure. Indeed, the development of the Welsh affricate subsystem 
(Griff en 197^b) can be interpreted as the replacement of this variable 
x by the symbol for the palatal -alveolar order /3/» borrowed from 
English. Likewise, the development of the voiceless palatal and bilabial 


fricatives (see section 7.2) can be interpreted as the replacement of 
this variable x by the symbols /y/ and /w/, respectively. Thus, we 
can extend the notion that the prosodic opposition of aspiration 
operates without regard to segmental boundaries to a stronger one which 
states that the opposition of aspiration, which forms the basis of the 
consonant subsystem, operates without regard to segments. 

That a consonant subsystem should be organized through a prosodic 
nonsegmental base brings into question the very notion of the segment 
in phonology. Indeed, if the basic opposition of our phonological 
consonant subsystem operates without regard to the boundaries of segments, 
how are we to define a segment phonologically? More pointedly, if the 
basis of our system does not need to rely upon the notion of a segment, 
why is our description couched in the terminology of a segmental phono- 
logy? It appears as though we may be unjustifiably restricting our 
phonology by demanding that the base of the structure depend upon some 
peripheral notion for its definitions. 

What we need to do now is to reexamine the notion of segmental 
phonology. We may be relying upon a segment which is merely some 
fiction introduced for the convenience of description. If so, the 
segment would be a violation of the principles of abstraction, as it 
would introduce an element to one level of abstraction which is not 
present on the lower level. As the lower level of abstraction is 
phonetic, we must examine the phonetic data in order to see what justifica- 
tion there may be for the use of the notion of the segment. Such an 
examination is also necessary from the inner approach to phonology, 
which holds that the phonology must reflect the phonetic sounds of the 
language. This examination is conducted in the next chapter. 


Items from Fynes-Clinton 1913 have been slightly altered wherever 
necessary to bring them into line with the notation used herein. As 
in section 5.2, the symbol for the voiceless stop represents voiceless 
aspirated, that for the voiced stop represents voiced unaspirated, that 
for the voiced stop with a dot subscript represents voiceless unaspirated. 
A circle below any sonorant represents the voiceless cognate. 

2 In the Welsh phonological system, /s/ is taken to be aspirated. 
See Table 1.0. c. 

-7s/ does enter into relationships in some dialects. 

ys/ is not included as it does not enter into phonological 
relationships in Standard Welsh. It is dealt with in Chapter 9. 

5rhe numeral 2 in the citation refers to sentence intonation. 

°The lines connect the consonants displaying alliteration. The 
corresponding consonant sounds can be found in Tables 1.0. a-b. 

7/hy/ represents a palatal voiceless fricative and /hw/ represents 
a bilabial voiceless fricative (rounded), 

8As mentioned in section 1.1, the Old Welsh /y/, which would cor- 
respond to /gM/i and "the Old Welsh /f3/, which would correspond to the 
soft mutation of /bn/, were lost in Middle Welsh. Thus, the structure 
shows inconsistencies of pattern reflecting these historical changes. 



8.0 The Phonetic Basis of Oppositions . On the phonetically pertinent 
level of abstraction, as pointed out in section 7.0, the basic element 
is the opposition. The phonetic opposition occupies an important place 
in the process of abstraction leading to the development of a phono- 
logical structure. As is characteristic of all such processes of 
abstraction, this basic element enters into relationships of abstraction 
with the elements of the phonological level of abstraction (that is, 
the more abstract) and with the elements of the actual phonetic data 
(that is, the less abstract). Thus, in order to defend one or another 
phonological system on the basis of the acoustic and physiological 
data, it is necessary to operate through the phonetic opposition. 

The strategic location of the phonetic opposition in the overall 
structure of the phonology demands that the characteristics of the 
phonetic oppositions be especially defensible within the notions of 
abstraction. In order to maintain a high degree of defensibility, it 
is necessary to introduce a minimum of assumptions along with the 
abstraction process. The requisite of a minimum of assumptions in the 
determination of phonetic oppositions can be achieved, so long as those 
characteristics of the phonetic oppositions on the level of phonetic 
pertinence correspond to particular characteristics of the actual 
phonetic data, both acoustic and physiological, in a regular and 
predictable way. 



To be sure, there are assumptions that must be made in the process 
of abstraction from the phonetic data to the phonetically pertinent 
oppositions. The most glaring assumption is that acoustic and physi- 
ological characteristics constitute the essence through which speech 
is transmitted. Actually, this is a conclusion which follows from the 
lack of any viable alternative. 

Given the assumption that the acoustic and physiological charac- 
teristics form the phonetic data from which oppositions are abstracted, 
we must determine what characteristics to abstract in order to form 
particular oppositions. This is usually done by noting which charac- 
teristics recur without exception in the production and perception of 
certain sounds. This method can also be reinforced by methods of 
synthesis, in which the sounds are synthetically reproduced from 
devices making use of only the characteristics under consideration. 
Still, however, this process involves assumptions; namely, that it is 
a particular characteristic which recurs which is responsible for the 
sound, and not some characteristic of the characteristic. 

This latter assumption involves our present level of sophistication 
in the area of instrumental phonetics. Until our knowledge of phonetics 
is expanded, however, we must accept the assumptions of acoustic and 
physiological phonetics as they are now practiced, if we are going to 
attempt the construction of any phonology from the inner approach. On 
the other hand, we must always be prepared to abandon current assumptions 
whenever our knowledge of phonetics is expanded through experimental 
results and these assumptions are found to be unjustifiable. 

If the phonetic opposition is to be a defensible notion in the 
abstraction of a phonological structure, it must be correlated with some 


characteristics consistent Kith our present knowledge of phonetics 
which obtain on the lower level of abstraction. Each correlation 
between phonetic opposition and characteristic tends to support the 
opposition in its particularly strategic position in the structure. 
This supporting evidence can be found in the two established areas of 
instrumental phonetics— acoustic and physiological phonetics. 1 

Over the past several decades, acoustic phonetics has enjoyed a 
considerable popularity and has afforded significant insights in the 
field of phonetics. This has been accomplished primarily through the 
electrical analogs and the instruments produced through them, in par- 
ticular the sound spectrograph of Koenig, Dunn, and Lacy 19^. Of 
course, the basis of such systems is theoretical (see Fant i960), but 
the assumptions involved in the acoustic theory of speech production 
should be accepted by the linguist as an expression of the limits of 
our present knowledge of acoustic phonetics (though, to be sure, such 
acceptance should be wary). 

Much of the effort in acoustic phonetics expended on the problem 
of finding correlations between phonetic oppositions and particular 
characteristics has been spent in the determination of vowel sounds. 
Through spectrographs analysis, for example, it has been determined 
that the various vowel sounds correspond to formant frequency charac- 
teristics (see, for example, Peterson and Barney 1952). These charac- 
teristics form the basis of certain phonetic oppositions (Fant 1959i 
1962; Jakobson, Fant, and Halle 1952). 

Spectrographic analysis has also afforded significant insights 
into the determination of acoustic characteristics relating to the 
various consonantal oppositions. For example, such oppositions as 


closure and vocal onset time have been correlated with characteristics 
in the spectrograms of stops (Fischer-J^rgensen 195&J Halle, Hughes, 
and Radley 1957). Similarly, the various oppositions of fricatives 
(Strevens 1960j Heinz and Stevens 196l) and of glides (Lehiste and 
Peterson 196l) have also been correlated with characteristics found 
in the spectrogram. 

The determination of phonetic characteristics underlying con- 
sonant oppositions has been greatly aided by such instruments as the 
playback spectrograph, which allows the experimenter to manipulate these 
characteristics (see Cooper, Liberman, and Borst 196l). Through the 
methods of synthesis, phoneticians have been able to make a much 
stronger correlation between precise characteristics on the spectro- 
gram and the phonetic oppositions. Such characteristics include transi- 
tions which vary from consonant to consonant depending to a large extend 
on the vocalic environment (see Gooper, Delattre, Liberman, Borst, 
and Gerstman 1952 > Delattre, Liberman, and Gooper 1955 1 Ohman I966), 
I return to this notion of characteristics of consonants depending on 
those of vowels in the next section. 

At our present level of sophistication in acoustic phonetics, 
taking into account the theoretical nature of the analogs, we can find 
strong correlations between the phonetic oppositions and particular 
characteristics in the spectrogram, characteristics based upon frequency, 
time, and amplitude. The fact that these characteristics should main- 
tain a basis in observable and even measurable data is of especial 
importance in our construction of a phonology through abstraction. 
Through the use of these acoustic measurements, we can base our abstrac- 
tions upon concrete items; and as long as we do not violate any principles 


of abstraction, our oppositions have a high degree of reliability on 
account of these concrete items. 

The other well-established area of phonetics is physiological 
phonetics. This branch of phonetics is actually the older of the two, 
a fact reflecting its greater independence from instrumental methods. 
Physiological phonetics, as it is known today, began in the last century 
as articulatory phonetics, especially in the visible speech of A.M. Bell 
1899, A.G. Bell 1909, and Sweet 1909 (see also Sweet 188^ on Welsh). 
The notion of articulatory phonetics was used as a basis in the feature 
(opposition) systems in American structuralism (compare Bloomfield 
1933)t as well as most other schools of phonology (phonemics). 

More recently, the development of adequate x-ray techniques, and 
especially of cineradiography (see Shawm and Stevens 1963)» has expanded 
the notion of articulatory phonetics to the more general physiological 
phonetics. Moreover, in so doing we have acquired a basis for a 
physiological theory paralleling the acoustic theory of speech production 
(see Peterson and Shoup 1966). 

Such studies as Perkell I969 not only increase the viability of a 
physiological theory, but they supply observable phonetic evidence that 
can be accurately measured. For the purpose of abstracting phonetic 
oppositions from observable characteristics, this last point is extremely 
crucial. In Perkell 1969, for example, we find precise physiological 
characteristics, characteristics based upon space and time. As we find 
in the measurable acoustic characteristics, we can use these physiological 
characteristics for the construction of a phonology through the abstraction 
of oppositions from concrete items. 

Both the evidence from the spectrogram and the evidence from the 


cineradiogram give us precisely measurable, concrete items from which ve 
can abstract oppositions. Moreover, as these instrumental readings 
occur in a time-oriented manner, they give us the means of dynamically 
measuring the actual occurrences of speech, a point to which I return 
in the next section. 

I mention above that acoustic and physiological phonetics form 
the two established fields of phonetics. There is one other field of 
phonetics that ought to be noted. This is neurological phonetics. As 
speech is a function of the neural system (see, for example, Penfield 
and Roberts 1959) » we should also try to find neural characteristics that 
correlate with phonetic oppositions (compare Liebennan 1970). Although 
some work has been done in this area (for example, MacNellage 1970), the 
field has not yet developed to the point at which we could reliably 
abstract oppositions from neural characteristics. Nonetheless, we 
should be prepared to modify our present abstractions on the basis of 
new developments in this field (compare Reich 1968). 

Reviewing the literature of acoustic and physiological phonetics, 
we find ample evidence to support the notion that certain concrete, 
measurable characteristics occur in the actual phonetic data which we 
can use as a basis for abstracting phonetically pertinent oppositions. 
Oppositions abstracted from these data maintain a high degree of 
reliability, at the very least. This high degree of reliability should 
be reflected in the overall structure of our phonology, if we manage 
to restrict our abstraction process so as to allow a minimum of assump- 
tions in addition to these oppositions. 

The danger of introducing faulty assumptions does not lie in the 
abstraction of the phonetic oppositions (provided we maintain the 


correlation with phonetic characteristics), "but it lies in our organiza- 
tion of these oppositions into phonetic, and then phonological, struc- 
tures. Nonetheless, these oppositions must be organized into systems 
within a structure, for we could not possibly make inferences as to 
the nature of phonology (and language in general) if we were faced with 
nothing more than a conglomeration of phonetic oppositions in no par- 
ticular order. On the phonetically pertinent level of abstraction as 
well as on the phonologically pertinent level, the oppositions must 
enter into relations one with another, if we are to construct a func- 
tional structure of phonology. Only if these relations are established 
can we gain some insight into the language. 

The organization of oppositions into structures must be based upon 
some premises. Now there are two ways of forming a basis for the 
organization. On the one hand, we could organize the oppositions in 
accordance with notions that create a consistent structure but are not 
based upon any evidence derived from the observation of some natural 
relationships that appear to hold between the various phonetic charac- 
teristics themselves. Of course, if we cannot find any such evidence 
of natural relationships between phonetic characteristics, this is our 
only choice. If, on the other hand, there is some observable set of 
relationships that obtain between the various phonetic characteristics, 
then we are obliged by the inner approach to abstract the relationships 
and use them in organizing the higher levels of abstraction. 

8.1 The Organization of Oppositions . The usual manner of organizing 
oppositions in order to construct a phonological (or phonemic) structure 
is through the use of the segment. A segmental phonology is based upon 


the premise that sound is segment able. The various oppositions in a 
given sample of speech can be categorized as belonging to one or an- 
other unit of oppositions, each of which is completely included in a 
segment. Once all of the oppositions are organized into segments, we 
have a string of discrete segments, or phones. 

On the level of phonetic pertinence in such a framework, these 
segmental phones can be manipulated such that phones which include 
particular opposition members are grouped together. The usual way of 
grouping the phones is through the principles of noncontrastive dis- 
tribution. For example, in English the phone [t h ] occurs in initial 
position in tonic syllables and contains the aspirated member of the 
opposition of aspiration} while the phone [t] occurs in initial position 
in tonic syllables after [s] and with the unaspirated member of the 
opposition of aspiration. Likewise, from the evidence submitted in the 
previous chapter, we can group the Welsh phone [d]*ich may occur In 
word-final position with the voiceless member of the opposition of voice 
along with the phone [d] with the voiced member of the same opposition. 

Having organized the phones into groups of allophones based on 
the principles of noncontrastive distribution, we can abstract the 
phonologically pertinent oppositions, those oppositions which are shared 
by the various members of each group of allophones, in order to form 
the phonemes. These phonemes form the basic organizational unit of 
oppositions on the level of phonological pertinence. These phonemes, 
moreover, can further be grouped in the phonology according to the 
various opposition members contained in them. The usual way of doing 
this is through consideration of grammatical function, and the resulting 
units are known as morphophonemes, which form relations on a yet higher 


level of abstraction (or, In some schools, on an equal level with the 

This method of organization of the oppositions (or features) has 
a long history in structural linguistics. Indeed, the phoneme of de 
Saussure (1959 1 38-^9) is constructed along the lines of the abstraction 
process mentioned above, as is the phoneme as a 'bundle of distinctive 
features' in Bloomfield (1933 « Chapter 5). As we can see in the inherent 
oppositions of the Prague school (Trubetzkoy 19^9), the notion of the 
phonological segment, though far more abstract, is basically the same 
there. In more recent developments in phonology, the underlying segment 
(or representation) of the generative school of phonology (Chomsky and 
Halle I9681 Schane 1973) is firmly rooted in the segmental -phonemic 
tradition in spite of the problems involved in deriving continuous 
speech from these discrete segmental units (see Liberman 1970 j Liberman 
et al. I967). Although the stratlficational approach does not introduce 
the segmentation process quite as close to the phonetic data as do the 
other schools, nonetheless there Is a phonemic level on which oppositions 
are ultimately grouped into phonemes for relationships with higher 
strata (Lamb 1966a, 1966bj Lockwood 1972tChapter 6~though Sullivan 1975 
does make use of nonsegmental description for voice in Russian obstruents). 
Even in the system-structure approach of prosodic analysis (Firth 1°48| 
Robins 1957) in which the prosodic, nonsegmental opposition figures so 
prominently, there is still a segmental base in relation to which the 
prosodies operate. 

The history of the segment in structural phonology (or phonemics) 
has been long, and its influence has, if anything, become increasingly 
felt throughout linguistics, regardless of the particular theory. We 


might expect, then, that the process of segmentation would he fully 
justified in the construction of a phonological structure. As we see 
below, however, this is not the case. 

One of the basic facts of phonetics is that sound is not in reality 
segmented. That is to say that we do not hear a succession of discrete 
segments, but a continuum of sound. The process of segmentation involves 
a method of organization of oppositions which, to be sure, creates a 
consistent structure, while not basing this organization upon any evidence 
derived from the observation of any natural relationships that appear 
to hold between the various phonetic characteristics underlying the 
oppositions (compare the previous section and Fant 1962). 

The nature of speech sound as a continuum has not gone unrecognized 
in phonology, Twaddell 1935 recognizes the phoneme not as a real unit 
of sound, but as a convenient fiction. The concept of a segment being 
a unit of convenience rather than a real unit of sound is also recog- 
nized in Schane (1973 Chapter l)t although some psychological justifica- 
tion for the segment is offered, based on the nature of the western 

Indeed, phonologists generally recognize the fact that speech sound 
is not segmental, that the process of segmentation is merely a method 
of organization without any basis in the acoustic or the physiological 
data. Although the lack of phonetic justification in this process does 
call to question the role of segmentation, the use of segmentation has 
been necessary in the past because of a lack of any evidence of natural 
relationships that obtain between phonetic characteristics that could 
be used in the more abstract levels of the structure. Moreover, further 
use of the segment in phonology is still necessary unless such evidence 


is provided from the fields of phonetics. Again if such evidence is 
found, then we are obliged by the inner approach and by the notions of 
abstraction to abstract any relationships that obtain between the various 
characteristics of the phonetic data which can be used for the organiza- 
tion of phonetically (and phonologically) pertinent oppositions. This 
is to say that notcnly the oppositions themselves but the relationships 
between oppositions should be based upon phonetic evidence wherever 

I mention in the previous section that the evidence from the 
spectrogram and the cineradiogram affords measurable, concrete items 
from which we can abstract phonetic oppositions. This evidence, more- 
over, is a function of time, making possible dynamic measurements of 
the speech event. The development of dynamic measurements of observable 
phonetic events makes possible the construction of phonetic models for 
the prediction of those dynamic movements of the physiological apparatus 
and the formant characteristics upon which these measurements are based. 

The successful development of any such dynamic phonetic model 
constitutes an important contribution not only to the fields of phonetics, 
but also to the structure of the phonology, for if such a phonetic 
model is produced, it must have some form of phonetically- justifiable 
organization between the characteristics from which we abstract the 
oppositions. This phonetic organization would provide the observable 
set of relationships obtaining between the various phonetic characteristics 
needed for abstraction into the higher levels of abstraction. If such 
abstraction can be performed, we would no longer have to rely upon 
segmentation for the organization of oppositions. Instead, we could 
organize the oppositions in a dynamic manner consistent with the observable 


data, allowing ourselves a greater degree of correlation between the 
phonology and the data. 

Such a dynamic phonetic model is what we find in Mermelstein 1973. 
The model constructed by Mermelstein for the production of vowel-con- 
sonant-vowel utterances is physiological and is based upon measurements 
with relation to the jaw, hyoid, tongue body, tongue blade, lips, velum, 
and maxilla and rear pharyngeal wall. 

In this model, the midsagittal dimension is divided into thirty- 
four sections, and measurements are based upon distances and angles 
computed from the intersections of lines tangent with the above-mentioned 
stationary and moving points of reference. Given the target areas of 
the proposed vowels and consonant, the movements of the pertinent 
articulators can be computed by the model. 

As I mention below, Mermelstein ' s model is not in Itself a sur- 
prising development. Indeed, it is a logical step in a series of 
phonetic developments over the past decade. The main contribution 
of this model is its mathematical predictability. Although others have 
suggested the same notions shown by Mermelstein in a hypothetical form 
(for example, Curtis 195^) » it is with this model that the hypotheses 
have been solidified into predictable calculations that can be empirically 
tested. As such, then, it necessarily forms an important part of a 
physiological theory of speech production. 

Another important aspect of Mermelstein' s model is the notion of 
hierarchical application of articulatory movements. According to this 
dynamic model, the trajectory between the first and second vowel of 
a V(o)V utterance is plotted as a function of time. Given the particular 
trajectory of the vowels, the physiological characteristics of a given 


consonant will apply In a particular order of importance — the movements 
associated with particular characteristics of most importance to the 
consonant are applied first, and the rest in order, if possible in 
the trajectory as modified by the more important consonant movements. 
In such a hierarchy, for example, stops and nasals produced at the 
alveolar position of articulation have the tongue tip marked 1 (most 
important), the jaw opening and tongue body marked 2, and the lip height 
and lip protrusion marked 3 (least important— articulated only if 
possible) . 

Once again, it is the formalism of Mermelstein which is of greatest 
importance here in the development of a physiological model of articula- 
tion. The notion that not all of the articulators actually reach their tar- 
get areas given certain information about the vowels is a significant 
development. I should further stress that these orders of application as 
well as the movements of articulators are not simply predicted in prin- 
ciple, but are computed mathematically. The predictions of movements 
based upon information relating to the vowels is further enhanced in 
the model by methods of predicting movements of consonant articulators 
in consonant clusters. 

Mermelstein (l973»108i-2) bases this dynamic model on the following 
principles « 

(1) The midsagittal vocal -tract outline is modeled in terms 
of nine selected variables describing the position of the 
participating articulators. 

(2) Stationary vowels are represented in terms of four 
variables, two describing tongue-body position, one the 

jaw position, and one the lip position. Movement from vowel 
to vowel, expressed as changes in the variable values, is 
slow and precisely controlled. 

(3) Representation of consonants requires additional control 
of tongue- tip elevation, lip height, and velar opening. Tongue 
body or jaw closure is specified by the variables pertinent 

to vowels. 


(4) Consonants are not defined directly in terms of vari- 
able values but by constraints on articulator position relative 
to the fixed structures. Articulators independent of the 
specific constraints are free to take on positions independent 
of the consonant under production subject to the requirement 
that they do not otherwise constrict the vocal tract. 

(5) Stop consonants are released by rapid movement of the 
constricting articulator. 

The development of Mermelstein 's articulatory model is, as I 
mention above, not the work of one man alone, but the logical extension 
of a decade of work not only in physiological phonetics but also in 
acoustic phonetics. I group the developments into two areas: the 
development of the model and the determination of the consonant-vowel 

In the development of the model, Merraelstein has several predeces- 
sors i although, to be sure, these are mainly concerned with static, 
not dynamic models. For example Heinz and Stevens 1964 uses anatomical 
measurements to determine a tentative set of rules for computing the 
area functions corresponding to articulatory configurations. Like 
Mermelstein 's model, this model relies upon rules specified in terms of 
functions relating cross-sectional areas to linear dimension. It 
also incorporates the effects of jaw and velum positions. 

Coker and Fujimura I966 constucts a static geometric model which 
achieves •reasonable accuracy' for both vowels and consonants. It 
uses five variables, representing the tongue body and the palate-pharyn- 
geal wall as circles, a practice later adapted by Mermelstein. Mermel- 
stein's use of jaw and lip coordinates, moreover, can be traced to 
Mermelstein, Maeda, and Jujimura 1971. Furthermore, the work of 
Lindblom and Sundberg 1971 also figures prominently in the development 
of the current model. 


Many, If not most, of the phoneticians mentioned so far are 
mainly interested in the acoustic aspect of speech production or in 
the interrelationships between acoustic and physiological aspects. 
Progress on the more clearly physiological side includes the findings 
of Perkell 1969. Perkell's measurements from cineradiographic film 
are instrumental in the development of the current model, as are the 
measurements of Mermelstein et al. 1971 . 

Not all of the phoneticians who contributed to the development of 
the articulatory model are concerned with merely static models. For 
example, (Shman 1966 notes that it is necessary to take anticipation 
into consideration in the construction of any model (acoustic or 
physiological). Furthermore, Shman attempts to relate the overlapping 
movements of the articulators to separate neural instructions (compare 
IiAerman 1970). 

Another important contribution in this area is that of Atal and 
Hanauer 1971. They by-pass the study of spectra (Fourier analysis) 
and represent the speech waveform directly in terms of time-varying 
parameters related to the transfer function of the vocal tract and the 
characteristics of the source function. In so doing, they find that 
the linear predictability of the speech wave forms the basis of both 
the analysis and the synthesis procedures. This development of the 
time variable in the construction of a speech model is of obvious impor- 
tance to the type of model introduced by Mermelstein. 

The mathematical predictability of Mermelstein' s model is, of 
course, of great phonetic consequence. But it is perhaps of even 
greater importance to the entire study of phonology that Mermelstein 
should have based these mathematical calculations on the principle that 


consonants are constraints on vowels. This dependence relationship 
between consonants and vowels, however, is not new with Merraelstein, 
As with the nature of his model, this aspect of his work is also a 
logical extension of developments within the field. 

In acoustic phonetics, the dependence of consonants on vowels 
has long been known. The locus and transition studies conducted at 
Hasklns Laboratory, particularly Delattre et al. 1955. establish the fact 
that aspects of consonant transitions are affected to a great extent 
by the nature of the vowels in the environment. These variations are 
predictable, to some extent, as shown by Liberman, Ingemann, Lisker, 
Delattre, and Cooper 1959. 

Indeed, one of the most important findings of Chman I966 is that 
there is a large discrepancy in the realizations of various consonant 
cues. This discrepancy can be attributed to the surrounding vowels. 

A significant review of the literature in this area is found in 
Liberman et al. 196?. In addition to the review, Liberman et al. 
propose a model by which speech is coded and decoded (as opposed to 
being handled as a cipher) by intricate physiological -acoustic -neural 
mechanisms. Regarding the transmission of consonants and vowels, they 
propose that these phoneme types are transmitted in parallel, rather 
than in sequence, Liberman 1970 also presents this argument, basically 
from an acoustic standpoint. 

From a physiological standpoint, perhaps the most important 
influence upon the current model's approach to consonant-vowel rela- 
tionships is found in the work of Perkell 1969. According to Perkell 
(I969165-6), the effect of the vowel upon the consonant is one of 
position of articulation, rather than of manner of articulation. More- 
over, the deformation of the articulating organ in the production of 


the consonant is superimposed upon the vowel (the positioning element) 
through fast, precise intrinsic musculature. We can see a definite 
parallel between these findings from cineradiography and the principles 
upon which the current model is based. 

In this respect, Perkell 1969 is in full agreement with-Ohman 1967, 
in saying that the consonant can be described as being superimposed 
upon continuously varying vowel articulators, a notion central to the 
current model. Moreover, the findings of Ohman 196? not only agree 
with determinations in physiological phonetics, but they provide a 
basis for the extension of the same dynamic model into the realm of 
acoustic phonetics, 

As I state in the previous section, if there is some observable 
set of relationships that obtain between the various phonetic character- 
istics, then we are obliged by the inner approach to abstract the rela- 
tionships and use them in organizing the higher levels of abstraction. 
The evidence from physiological and acoustic phonetics will not support 
an organization of oppositions in the phonology based upon the notions 
of segmentation, but it will support an organization of oppositions 
based upon principles of hierarchical application (compare section 7.1). 
Thus, the demands of the inner approach to phonology leave us no other 
choice than to abandon the abstraction into segments in favor of some 
sort of abstraction into a hierarchy of oppositions. 

The demands of the inner approach are not merely conventional 
devices which can be modified for the sake of convenience and simplicity. 
Rather, the inner approach is a statement of theory saying that phonological 
structures must be abstracted from real phonetic evidence. Once we are 
faced with real phonetic evidence which can be abstracted to form 


phonological relationships, If we choose to avoid this evidence, we 
are no longer working within the same theory. More importantly, if 
we choose to avoid this evidence, we are basing our structure not upon 
fact, but upon fiction. 

The choice, then, between incorporating our new-found evidence 
from physiological and acoustic phonetics or ignoring it in favor of 
the traditional segmental -phonemic approach constitutes a choice in 
the direction which phonology is to take in the future. On the one 
hand, we can choose to incorporate the organizational structure found 
in the phonetic evidence, in which case we will be faced with the 
formidable task of constructing a phonology through some extremely 
complex relationships. But, however long and however difficult this 
task, we will be assured that our phonology has a sound, real basis in 
observable evidence which will afford us insights into the nature of 
language. On the other hand, we can choose to avoid the evidence and 
maintain a segmental approach to phonological organization, in which 
case our investigations in phonology will be much simpler and easy to 
understand. Of course, along with this simplicity and ease we will 
also realize that we are working in a fictional realm which cannot 
afford us insights into the nature of language, but only affords us 
insights into the nature of the particular model used in description 
(compare Chapter 6). This is not to say that investigations in the 
segmental framework cannot be useful In directing our attention to 
other areas of investigation (indeed, we see this in the phonological 
evidence of the previous chapter). Ihat I am saying is that investiga- 
tion in the segmental mode, while occasionally useful, ought not to be 
confused with investigations relating directly to the sound systems of 


real languages. 

Most phonologists have realized that the relationships obtained 
through the segmentation of speech sound do not constitute real evidence. 
Nonetheless, the segmentation process has been necessary in the past 
because of the absence of any phonetically justifiable manner of 
organizing the oppositions into a viable phonological structure. This 
situation, however, has now been eliminated by the development of 
phonetic models on an hierarchical basis. It is now the task of the 
phonologist to formulate a phonological structure on the basis of these 
hierarchical relationships. 

8.2 An Hierarchical Structure in Phonology . In this section I set 
forth some initial definitions and postulates which may be useful in 
the construction of a phonological structure along the lines of the 
phonetic relationships outlined in the previous section and in section 
7.1. Such a structure I term an hierarchical phonology. These pos- 
tulates and definitions are consistent with the findings of Mermel- 
stein 1973 and Perkell 19&9 in physiological phonetics and with the 
findings of (5hman 1966 and 1967 in acoustic phonetics. Moreover, they 
are consistent with the general phonological theory involving abstraction 
into levels of pertinence which I have been using in this and the 
previous chapter. 

In stating these postulates and definitions, I use the familiar 
formula *x is y_*. This is certainly not to be read as an irrefutable 
observation of tangible phenomena, but should be read 'within this 
approach, we consider x to be y_' , Subsequent research may bring about 
changes anywhere in this structure? nor should such changes be avoided, 
so long as these changes are based upon evidence. 


The model suggested In these postulates and definitions is based 
upon the notion (supported "by the phonetic dynamic models) that con- 
sonants are constraints on vowels. This is to say that the vocalic 
oppositions are realized in a continuous stream, and various types of 
consonantal oppositions obstruct the flow of this stream. We my thus 
consider such notions as a consonantal opposition member obscuring a 
vocalic one such that the vocalic opposition member is acoustically 
altered in some way (is no longer 'clear'). In such a system, con- 
sonants cannot exist separate from vowels. 

The vocalic pattern is a continuous stretch of speech from one 
vowel approximation to the next, interrupted only by constraints on 
the vowels themselves. Physiologically, this represents continuous 
movement of variables relating to tongue-body position, jaw position, 
and lip protrusion. So long as the movement of these varia b les is not 
altogether terminated in the speech function (that is, in communication), 
one vocalic pattern is maintained. 

Acoustically, the steady movement of vowel -determining articulators 
is representable in terms of formant continuity. If the formants 
on a spectrogram are either continuous or their gaps, produced by 
constraint rather than termination or total interruption of the speech 
function, can be interpolated, then one vocalic pattern is maintained, 

A vocalic pattern nay be as short as, but no shorter than, a sus- 
tainable sound the size of a syllable or as long as a breath group 
(compare Lieberman 1967), or, indeed, as a discourse block. It is not 
a subset of a reasonably determinate set of possible vocalic patterns 
(although, to be sure, the number is finite). 

A syllable is a member (or unitary member) of a vocalic pattern. 


It is characterized by a vowel approximation. In physiological terms, 
this vowel approximation corresponds to the target area of a vowel 
at which a steady state harmonic sound could be emitted. In acoustic 
terms, it ism area of the spectrogram in which forraant frequencies are 
definable by poles and zeros (see Fant 195&), and in which a steady-state 
set of frequencies is approximated (though not necessarily achieved). 
Although the syllable cannot be obscured physiologically (that is, the 
vocalic articulators always maintain their positions regardless of 
consonants), it can be effectively obscured by obstruction acoustically 
(such as the Mandarin syllable G9 [sz] 'four', which has no 'pure' 
unobscured vowel sound). Nonetheless, some formant patterns can be 
discerned in all syllables. A syllable may be the shortest sequential 
unit (see Malmberg 1955; Liberman et al. 196?). 

The vocalic oppositions are measurements of physiological and 
acoustic variables. Whether physiological or acoustic variables are 
used in any description is not a matter of one's supremacy over the 
other, but reflects rather the demands of the particular description 
(Lieberman 1970). Indeed, the choice ought not to be an exclusive one. 

As the vocalic pattern is traced from syllable to syllable, 
movement from one vowel approximation to the next will cause functional 
movement in the physiological vocalic variable and in the acoustic 
formant patterns. These transitions are relatively slow and form the 
boundary areas of syllables. Quicker transitions are not syllable 
boundaries but elements of obstruction or restrictions and are associated 
traditionally with consonants and glides (Lehiste and Peterson 1961). 

A vocalic restriction (associated with glides) is a medial transition 
(faster than a syllable boundary but slower than an obstruction element^ 


It commonly accompanies a length of reasonably steady vocalic emit ion, 
but It is too short itself to maintain a steady state. As the vocalic 
restriction is medial in nature, a change in tempo (deliberate or 
Incidental) can cause the restriction to be alternatively realized 
as a syllable, a syllabic boundary (if it corresponds to the following 
syllable), or an obstruction element. This is demonstrated in Lehiste 
and Peterson 1961. 

Obstructions vary according to language and, conceivably, to 
idiolect. Basically, an obstruction is a constraint on the vocalic 
pattern that occurs intersyllabically. During the obstruction, the 
physiological vocalic variables may be slightly constrained, but they 
continue in the vocalic pattern, and acoustically the transition from 
syllable to syllable can be interpolated mathematically (Ohman 1967). 
The obstruction is characterized (that is, the basis of the opposition 
is identified) as a particular position of articulation or acoustic 
locus . ' 

In some languages, there may be degrees of obstruction, such as 
total obstruction (1st degree— closure), partial obstruction (2nd degree— 
frication), and minor obstruction (3rd degree— sonorance) . Other 
languages, however, may operate in a way such that some of these degrees 
of obstruction are determined by prosody (as I show for Welsh in the 
next chapter). Lower degree obstructions may last for entire syllables 
(such as the Mandarin word cited above). Just what is a lower degree 
obstruction, of course, is dependent upon the structure of the particular 

Obstructions are not consonants in the segmental sense, there 
being no such thing in this nonsegmental phonology. They are particular 


constraints on the vocalic pattern that occur not sequentially with 
the vocalic pattern but concurrently with it and superimposed upon 
it. However, they are in sequence with each other. Thus, the nature 
of the vocalic variables in the syllable transition will determine to 
a great extent the precise manner of occurrence of the obstruction 
(as it lacks its own base for articulation), and it will, by virtue of 
its demands on articulators and acoustic locus requirements, affect the 
nature of the vowel upon which it is imposed in a predictable (phonetic) 
manner, but in a Banner of less magnitude than that of the vowel on the 
obstruction. Of course, the notion of allophonic (combinatory) variation 
becomes an hierarchical matter rather than a sequential one and is to 
be handled phonetically (that is, it is calculable from the formulae 
of Merraelstein 1973, Ohman 196?, etc.). 

In addition to the vocalic pattern, syllable, and obstruction, 
there is a series of prosodies. A vocalic pattern prosody includes 
the notion of intonation and consists of fluctuations in stress and 
pitch, among other things (see below). The actual position of occur- 
rence of these fluctuations will depend upon prosodies of other levels 
of the structure, notably those in words (which I do not define). 

A syllable prosody is a realization of a particular quality for 
the duration of a syllable. Such prosodies include tones, tunes, 
pitch, and stress. As the basis of the syllable is vocalic, there is 
no issue as to whether syllable prosodies occur on the nucleus alone. 
Such prosodies occur in the syllable and may be obscured by an obstruc- 
tion of a high degree (1st degree and possibly 2nd degree, depending 
upon the language). 

Vowel oppositions are, by the nature of the syllable, syllabic 


prosodies, such as height and frontness (physiologically) or compact- 
ness and gravity (acoustically). These prosodies can affect or alter 
those of dependent syllables in such phenomena as umlauting (vowel 
affection) and harmony. This is accomplished by the dominant syllable 
prosodies (vocalic oppositions) altering transitions which, in turn, 
affect adjoining syllable prosodies. (Note that the method is continu- 
ous — there is no hopping of segments by oppositions.) 

The most intricate set of prosodies is the set of obstruction 
prosodies (compare Trubetzkoy 1969ilk6), Like other prosodies and 
obstructions, the obstruction prosody is superimposed upon the vocalic 
pattern and is, as such, a constraint. These prosodies are associated 
with the obstructions, however, and may dominate part or all of the 

Obstruction prosodies can include tension, aspiration (the inverse 
of voice — voice is, after all, an inherent quality of the vocalic 
pattern and cannot be a constraint upon itself), nasality, and so forth. 
In some languages, as mentioned above, the obstruction prosody of 
tension, or prosodic tension, can determine the difference between 
stop and fricative (as in Welsh — see Griff en In press a; sections 6.1 
and ?,2, above). 

More than one obstruction prosody can occur at one time. For 

example, the Welsh nasal affricate in /nhain/ '(my) chain' Includes an 


apico-palatal -alveolar obstruction, a prosodic aspiration that is 
initiated concurrently with the onset of obstruction and dominates 
beyond the release of obstruction, and prosodic nasality that will vary 
considerably awng speakers and dialects, conceivably lasting into the 
second obstruction (dominating the entire syllable). Moreover, an 


obstruction prosody need not occur In conjunction with any (phonological) 
obstruction, as in French un /%/ 'one', and in the 'glottal fricative' 
/h/ (preaspiration). 

In cases such as whispered or nasalized speech (where it is con- 
sidered phonetically or phonologically pertinent), the obstruction 
prosody is superimposed upon the entire vocalic pattern. Thus, it is 
classified as a vocalic pattern prosody in this instance. Where nasality, 
for example , is a vocalic pattern prosody, it can be in addition imposed 
as an obstruction prosody (such as in nasal speech). Whispering is 
prosodic aspiration superimposed upon the vocalic pattern. 

The final opposition is the notion of speech itself, the weaker 
member of which is the absence of speech — breath. Breath is charac- 
terized by aspiration (voicelessness) of a lenis and possibly nasal 
manner (although, for purposes of markedness, it is considered to be 
nonnasal) in the absence of a vowel. Hence, it is prosodic without 
obstruction. Unlike whispering, breathing is characterized by a 
steady physiological state with articulatory apparatus in neutral 
position (the /h/ of Gleason 1961; or the voiceless schwa). Of course, 
the absence of the vocalic pattern in breath defines the lack of speech. 

Because of its prosodic nature, in languages in which tension is 
contrastive (possibly through aspiration, as in Welsh), breath can 
spread prosodic lenition (which may involve 'voice'— the reduction of 
aspiration) over final obstruents in anticipation of the lenis nature 
of breath j while in languages in which 'voice' (that is, lack of aspira- 
tion) is contrastive, it can spread prosodic aspiration over final 
obstruents. Thus, we find the peculiar situation of archiphonemic 
neutralization of 'voice' in word-final position in German and Welsh 


being realized in opposite ways (see sections 6,2 and 7.0), 

All restrictions, obstructions, and prosodies are realized by 
degree. The degree of realization, or constraint, is determined by 
such considerations as the rate of speech as well as by predictable 
considerations dictated by the environments. Of these predictable 
considerations, the type of hierarchy shown by Mermelstein (in 'marking' 
articulators) figures prominently. 

The division between phonology and phonetics in this nonsegmental , 
hierarchical system is, of course, not a matter of segmental versus 
nonsegmental. In this system, mathematically predictable and calculable 
values such as the exact range of a prosody or the precise degree of 
an obstruction become the realm of phonetics, while functionally con- 
trastive combinations and extents of constraints and vocalic values 
become the realm of phonology. Moreover, the difference between an 
opposition of phonetic pertinence and one of phonological pertinence 
is still a matter of contrast and predictability, as outlined in 
section 7,0, 

Using these postulates and definitions, we should be able to 
construct an hierarchical phonological structure. The details of the 
phonological structure of a particular language depend, in the final 
analysis, on the particular oppositions found in the data of that 
language. Thus, although the principles of abstraction of oppositions 
and relationships inherent in the postulates and definitions are general, 
the resulting structure will be particular. 

8,3 Toward an Hierarchical Phonology . A theory of language (or of 
some part of language) consists of certain principles concerning the 


nature of language. Through these principles, the linguist forms 
hypotheses and checks these hypotheses against the theory (for consis- 
tence) as well as against the data of the object of study. Hypotheses 
which adhere to the observations and to the theory are then formed into 
a model, from which the linguist constructs grammars. 

The theory of hierarchical phonology states that the process of 
abstraction into levels of pertinence can be accomplished from phonetic 
characteristics in order to construct a phonological structure. As 
such, then, the theory is by no means new. In fact, it is nothing more 
than the traditional theory of phonology through the inner approach 
(Jakobson and Halle 1962j Trubetzkoy 1969)# The postulates and 
definitions of the previous section constitute the hypotheses consistent 
with the theory and (assuming the accuracy of phonetic evidence) with 
the observations. From these hypotheses, we can construct a model, or 
a phonology. By applying this model to the data from a particular 
language, we can construct a grammar of the phonology of that language 
(as in the next chapter). 

As the theory of which hierarchical phonology is a model is the 
same theory that has traditionally been used with the inner approach 
(a combination of abstraction theory and opposition theory making up 
the functional/structural approach of Prague), and as my methods of 
using the theory in the construction of a model and a grammar are 
likewise traditional and (supposedly) known to all linguists, why, 
one might ask, do I go through the exercise of repeating them here? 
The reason for this lies in the confusion between model and theory. 
In recent years, it has increasingly become the practice of some 
linguists to construct a particular model within a particular framework 


(that is, adhering to certain hypotheses of organization) and then to 
defend this model as a theory, such that an attack upon the model is 
construed as an attack upon the theory (compare Chomsky 1965»2*0. 

I wish to avoid this misconception that an attack upon a model 
is an attack upon the theory, a misconception predicated upon the 
notion that they are isomorphic one with the other. Two decades ago, 
Hockett 195 1 * pointed out that three frameworks, three entirely different 
models, could be constructed from the same structural theory. Indeed, 
since that time, linguists using one or another model have attacked the 
other models of the same theory, acting as if they believed their attacks 
to be directed against some alien theory. The most harmful development 
has not been in the attacks (which are always beneficial, so long as 
they are to the point), but in the defenses, in which the proponents 
of one model or another defend even the most indefensible aspects of 
their model, believing (apparently) that this is an action necessary 
to save a theory of known value. 

Considering the history of the model -as-theory misconception, I 
should like to emphasize the relationship between the theory and a 
model. The model which I construct in this section on the basis of the 
hypotheses of the previous section is, I believe, more consistent with 
the theory and the observations than most (if not all) other models 
within this functional/structural theory. If, however, any aspect of 
this model or the model itself can be shown to be less consistent with 
the data and the theory than some other aspect or, indeed, some other 
model entirely, then I shall not defend the particular weak aspect or 
the model. In so doing, I defend the theory. On the other hand, a 
refusal to prune the model wherever necessary actually constitutes an 


attack upon the theory, albeit unwitting. 

The dispensability of a model is a logical necessity of linguistics. 
The reason for this is the abduct ive nature of model -construction 
(compare Anttila 1972iSection 9.16$ H. Anderson 1973). Within a logical 
syllogism, the major premise is the theory, the minor premise is the 
model, and the conclusion is the observation. Given a particular theory 
and a set of observations, we have the major premise and the conclusion. 
We now must determine the model, or minor premise. This process of 
abduction is by far the weakest logical operation we can conduct, for 
a given observation can be accounted for from a given theory by any 
number of models, few of which will be sufficient in the final analysis 
in their original forms. We must, therefore, maintain a high degree of 
flexibility in our construction of models and take special pain to 
eliminate all inconsistencies with the theory and the observations. 

The model which I propose from the postulates and definitions of 
the previous section (the hypotheses) is the hierarchy shown in Table 
8.3, This model represents a particular structure of phonology consistent 
(tentatively) with the theory and the observations. 

In this model, the oppositions in the higher levels are super- 
imposed upon those in the lower levels. The structure is divided into 
three general divisions. The vocalic pattern division provides the basis 
for the representation of the pertinent level (phonetic or phonological) 
in the abstraction of the speech event. The syllabic division serves 
to organize the vocalic pattern into sequential order and forms a basis 
for the obstruction division,^ The obstruction division constrains the 
vocalic pattern in accordance with the sequential order of the syllabic 










































































































Basically the same organizational structure can be used for the 
phonetic and the phonological levels of pertinence. The difference 
between the two, as stated above, is one of predictability versus 

In the construction of a grammar for the phonological system of 
a particular language, we organize the oppositions in accordance with 
these various divisions and levels. In describing different languages, 
we will probably have to adjust the basic model, and such an adjustment 
could serve as a statement of the organization of the particular language. 
In the next chapter, I describe the consonant system of Welsh within 
this model. 


1 The following excursus is a very basic survey of the literature 
and development of the field of phonetics. It is provided for the 
convenience of those readers who have not kept up with the literature, 

*0f course, the alphabet was invented only once (compare Gelb I963), 

3Position of articulation implies tongue configuration in addition 
to place of articulation, 

^There may be redundancy in these two divisions, a matter which 
will have to be determined in a later study of vowel phenomena. 



9.0 Theoretical Implications In the Application of the Model. As I 
point out in the previous section, a model of a particular theory con- 
sists of a set of hypotheses which we use to relate the theory to the 
observations. If we consider the theory to be the major premise and 
the observations to be the conclusion, the abduction of a model from 
the hypotheses requires a construction which satisfies elements of the 
theory as well as elements of the observations. 

The hypotheses and model considered in the previous chapter 
satisfy (tentatively) the elements of theory as well as observations. 
Turning to the sounds of a particular language, however, we find an 
entirely different logical operation. We must accept, at least tem- 
porarily, the model and the theory — in logical terms, the two premises. 
In this operation, we apply the model to the abstracted oppositions 
of the particular language in an attempt to deduce the observations. 
If the deduced observations correspond to the actual behavior of the 
language, then we can consider our phonological structure to be valid. 
If this correspondence is not found, then the structure is invalid 
and must be changed. 

The process of applying the model to a particular language involves, 
then, the logical operation of deduction. This operation involves a 
development of the argument from the theory and through the model, in 
order to arrive at specific observations. Such a process involves the 



development from general to specific. 

Thus, we find a process whereby deduction verifies abduction. 
This is not, however, all that is Involved. Before we can arrive at 
a theory, we have particular observations, and these observations are 
made with respect to a primitive model, which reflects those aspects 
of the observations which we consider, for one reason or another, to 
be important. The construction of a theory, then, involves the forma- 
tion of a major premise in a syllogism from the minor premise (the 
primitive model) and the conclusion (the observations). This process 
is induction. 

The formulation of a linguistic structure is subject to all three 
logical operations — induction, abduction, and deduction. These opera- 
tions, moreover, must be applied continuously, for each new development 
must be reconciled within the entire structure (compare Mulder 1 9681 1-7). 
Once overy element of the structure is logically defensible, then the 
entire structure is valid. Of course, validity ought not to be taken 
as the 'discovery of truth' (compare Robins 195?» also Chapter 6, above). 

The dependence of the three parts of a linguistic structure, the 
theory, the model, and the application of the model (the grammar), 
upon the relationships represented by the three logical operations 
should not be taken lightly. In spite of claims made by one school of 
linguistics or another, it is no more possible to create a linguistic 
structure purely from deduction (a theory created without considering 
any observations) than it is to create one purely from induction 
(gathering data with no idea of what it is that is being gathered). 
This point is an important one to remember, as we attempt to apply the 
model to a particular language in order to construct a grammar consistent 


with observations, for if, in the deductive process, the structure 
falters, then we should not hesitate to reapply the operations of 
abduction or even of induction in order to rectify the structure. 
Although we cannot change the observations and we avoid changing the 
theory, the model and its application in the grammar must remain 

Having considered the relationships that obtain between the various 
parts of the structure, we have yet to address the practical side of 
the problem — the construction of the grammar. The grammar is the 
application of the model to the particular language. In this case, we 
are concerned with the application of hierarchical model (Table 8.3) 
to the consonant (or obstruction) oppositions of Welsh. 

As noted in the previous chapter, the oppositions are abstractions 
from the phonetic data. Without some sort of organization, however, 
these oppositions can tell us little concerning the structure of the 
phonology. It is the model which supplies this necessary element of 
organization. By applying the model, then, we simply mean that we 
organize the oppositions into a grammar. 

The result of the application of the model is a functional arrange- 
ment of oppositions. This functional arrangement is peculiar to the 
particular language, for no two languages will have exactly the same 
oppositions and opposition members and exactly the same relationships 
holding between the various oppositions and opposition members. This 
peculiar functional arrangement of oppositions comprises the phonology 
of the language. 

There are two senses of the word 'phonology'. One pertains to 
the structure of the sound system of one particular language. This 


phonology is represented in the grammar of the language. On the other 
hand, phonology can also be used to designate the general notions 
embodied in the model. Thus, the phonology of Welsh, for example, 
adheres to the requisites of a general phonology. 

The relationship between the two notions of phonology raises 
the subject of the general versus the specific, or the universal versus 
the nominal. Although much attention has been given to this subject, 
as far as the practical problem of applying the model is concerned, 
it does not matter whether we consider the universal to be the common 
abstraction from the nominals or whether we maintain an ideal universal 
toward which all nominals strive. This subject is a philosophical one, 
to be argued as a matter of faith, and ought not to interfere with 
scientific investigation. As mentioned above, we ought not to limit 
ourselves to investigation from the general to the specific (deduction) 
or from specific to general (induction), but we should use all logical 
operations in order to establish a valid structure without philosophical 

In applying the model for the construction of a grammar, there is 
one other question which, in the light of recent linguistic practices, 
should be addressed. How formal and explicit should the structure be? 
When the structure is complete and the relationships have been established, 
some explicit formalism will suggest itself. To burden the linguist 
prematurely with an explicit set of formalism, however, would be to 
remove much of the needed flexibility from the linguist's operations. 
Such reliance upon a preconceived set of formalisms would suggest a 
final structure which may very well be inadequate. The formalism used 
herein, then, should not be taken as a necessary part of the notion of 


hierarchical phonology, but merely as an aid to conceptualization, 
to be changed or discarded at will, 

9.1 The Welsh Consonant Subsystem , In applying the hierarchical model 
to the Welsh consonant subsystem, we find two groups of consonant opposi- 
tions. First, we have the obstructions, which directly constrain the 
syllabic oppositions and the vocalic oppositions proper. Second, we 
have the obstruction prosodies, which are superimposed upon the obstruc- 
tions or upon their areas of constraint in the syllable transitions. 

As mentioned in section 8,2, the obstruction can be classified 
by degree depending upon the nature of phonetic realization. In Welsh, we 
can further classify the obstructions by the extent to which these 
degrees are realized through interaction between the obstruction and 
the prosodies. The primary obstruction is that which can be described 
by point of articulation only. It is the only one which maintains a 
realization of the 1st degree obstruction and enters into a full set 
of relationships with the obstruction prosodies. 

As with all obstructions, the primary obstructions are members 
of a gradual opposition of position of articulation. This position of 
articulation is a function of the point of articulation and the articu- 
lator configuration, and it can be determined both physiologically and 
acoustically. Since in Welsh there is no archiphonemic neutralization 
within the opposition of position of articulation, a determination of 
degrees of strength in the opposition cannot be made on the basis of 
neutralization. Nor, indeed, can such a determination be made on the 
basis of naturalness (in the manner of Trubetzkoy 19^9) » as all positions 
of articulation involve deviation from the state of breathing of 


approximately equal degree. 

Within the primary obstructions, there are (in Standard Welsh) 
three members of the opposition. These are the labial (the bilabial 
and the labio-dental), the (apico-)dental, and the frorso-) velar. 
Just for the sake of conceptualization, I use the following notations 
to represent these opposition members t /b/, /d/, /g/, respectively. 
As I state in the previous section, the notation is not to be taken 
as restrictive and is basically unimportant. 

The secondary obstructions are more restricted than the primary 
obstructions. They cannot be described by point of articulation alone, 
but rely also upon articulator configuration. Moreover, they do not 
maintain a 1st degree obstruction, and they are not as fully developed 
in their relationships with the obstruction prosodies (see Table 9.1.b). 
Within the secondary obstructions, there are two members of the opposi- 
tion. These are the lateral and the trill, for which I use (again, just 
for the sake of conceptualization) the notations /*/ and /*/» respectively. 

Finally, there is the tertiary obstruction. This is the most 
restricted classification. Like the secondary obstruction, it requires 
description by both point of articulation and articulator configuration 
(the full function of position of articulation) and maintains no obstruc- 
tion of the 1st degree. Unlike the secondary obstruction, it enters 
into only one relationship with one obstruction prosody (the minimum for 
articulation). In Standard Welsh, the only opposition member of the 
opposition of position of articulation in the class of tertiary obstruc- 
tions is the 'apico-alveolar slit fricative', designated by the notation 

A/. 1 

In the opposition of position of articulation, there is only one 


case of structural discrepancy between the phonetic and the phonological 
levels of abstraction. This discrepancy is found in the labial member 
of the opposition (/b/). Phonetically, the 1st degree obstructions are 
bilabial and the 2nd degree obstructions are labio-dental in Welsh. The 
fact that there are no conflicts between the two degrees (that is, there 
are no 1st degree labio-dentals or 2nd degree bilabials) shows that the 
difference between the two points of articulation is not contrastive. 
Thus, phonologically both degrees can simply be considered as labial 
(compare the principles of noncontrastive distribution). 

Superimposed upon the obstruction or directly upon the syllable 
transition is a class of oppositions known as the obstruction prosodies. 
These prosodies are not inherent in the obstructions, nor do they form 
some larger units with the obstructions (compare Chapter 7), In Standard 
Welsh, there are two obstruction prosodies — nasality and aspiration. 

In Welsh, the prosodic opposition of nasality is privative. The 
notation /n/ represents the nasal member of the opposition, and the 
nonnasal member is not overtly indicated. If the nasal member of the 
opposition of nasality is present, the velum is lowered allowing air 
to pass into the nasal cavity, resulting in the characteristic distribu- 
tion of acoustic energy. If, on the other hand, the nonnasal member of 
the opposition is present, the velum is not lowered and the characteristic 
distribution of acoustic energy is not effected. As an obstruction 
prosody, pertinent nasality is not affected by nasality as a vocalic 
pattern prosody, such as one finds in the nasalized speech of the 
Caernarvon district. 

While saying that the opposition of nasality is privative, I 
recognize that it is possible to perceive in Welsh some differences in 


degree of nasality, particularly in conjunction with aspiration. I 
would maintain, however, that this is not a result of a gradual opposi- 
tion of nasality, but merely the effect of aspiration operating in 
conjunction with it. 

The prosodic opposition of aspiration in Welsh is treated in some 
detail in Chapter 7, and I do not want to repeat the detail of that 
chapter here. Suffice it to say, however, that the opposition of 
aspiration is a gradual opposition, realized in four degrees of strength 
(or presence). For the sake of convenience, I represent the first 
degree (the weakest) with the notation /M/, the second degree with the 
absence of notation (or /#/), the third degree with the notation /h/, 
and the fourth degree with the notation /hh/. As always, I choose the 
particular notations simply because they are convenient to the subject 
which I am discussing (that is, historical changes in the Welsh language 
to the 'voiced unaspirated stop'). I could just as easily represent 
the opposition members as /lh/, /2h/, /3h/, and /4h/« 

In the structure of Welsh phonology, the prosodic opposition of 
aspiration is of prime importance, as evidenced in Chapter 7. Because 
of the importance of this opposition and of its gradual nature, I 
treat the phonetic evidence supporting the opposition in Chapter 10. 

The two groups of oppositions (obstructions and prosodies) in the 
obstruction division (that is the consonant subsystem) do not simply 
coexist. They enter into relationships in order to produce the various 
•sounds' of Welsh. These relationships consist of one member of the 
obstruction opposition and one member of either prosodic opposition or 
of both prosodic oppositions. These relationships can be economically 
described through the use of the three degrees of obstruction, mentioned 


The nasal member of the prosodic oppostion of nasality may enter 
into a relationship with any of the three primary obstructions, but 
there is a restriction on the relationships with primary obstructions 
into which the nasal member may enter. Nasality can only be realized, 
relative to a primary obstruction of the 1st degree. In traditional 
terminology, nasals in Modern Welsh can only be cognate with 'stops'. 
Moreover, nasality in Welsh can only be realized with an obstruction 
(though, to be sure, such realizations are possible in other languages, 
such as French and Portuguese — compare section 8,2). Of course, this 
restriction does not apply to Old Welsh, insofar as the realization of 
the nasal opposition member in a relationship with an obstruction of 
the 2nd degree is concerned, as we do find the bilabial nasal fricative 
in positions of soft mutation (compare Morris Jones 1913 »163). 

The relationships into which the various members of the prosodic 
opposition of aspiration may enter are far more complex than those of 
the nasal opposition member. All four members of the opposition of 
aspiration may enter into relationships with any of the three primary 
obstructions. When the first (or weakest) member of the opposition of 
aspiration enters into a relationship with an obstruction, the result 
is described in traditional terms as a 'voiced fricative'; when the 
second member enters into a relationship with an obstruction, the result 
is described traditionally as a 'voiced unaspirated stop'; when the third 
member of the opposition enters into a relationship with an obstruction, 
the result is described traditionally as a 'voiceless aspirated stop'; 
and when the fourth (strongest) member enters into a relationship with 
an obstruction, the result is described traditionally as a 'voiceless 
fricative'. Again, phonetic justification for these relationships is 


given in the next chapter. 

There is one apparent restriction in these relationships between 
members of the opposition of aspiration and the primary obstructions. 
When the weakest member of the opposition of aspiration enters into a 
relationship with the velar member of the obstruction opposition, 
phonetically the realization is null. Phonologically, however, the 
realization may be described through the notation /gM/. This acts as 
a place-holder in the phonological structure, and its identity is 
functionally important, especially as we consider the system of alter- 
nations in the next section. 

As demonstrated in section 7.2, the first (weakest) member of the 
prosodic opposition of aspiration may enter into a relationship with the 
secondary obstructions. The realization of this relationship in 
traditional terms is a set of liquids'. 2 The relationship between the 
second member of the opposition of aspiration and the secondary obstruc- 
tions is open to some interpretation. Based on the evidence of Chapter 
7, I have maintained this relationship in order to establish what is 
traditionally known as the 'voiceless lateral and trill fricatives'. 

On the other hand, one can also find evidence to support the 
notion that /l/ and /»/ are the realizations of a relationship between 
a higher member of the opposition of aspiration and the secondary 
obstruction*. First of all, there is the fricative nature of the rela- 
tionship, suggesting the fourth opposition member. Moreover, in some 
dialects, there have been reports of an aspirated lateral liquid as 
aspirate mutation form for the lateral liquid (Professor Ceinwen 
Thomas, University of Wales at Cardiff, personal communication, notes 
this in informants from Rhoshirwaun and Llannerchymedd, both in Gwynedd). 


In these dialects, at least, we would find a more complex set of 

As noted above, there is only one relationship between the opposi- 
tion of aspiration and the tertiary obstruction. The second member of 
the prosody enters into a relationship with the tertiary obstruction, 
such that the realization of this relationship can be described in 
traditional terms as the 'voiceless slit fricative' or the 'voiceless 
sibilant'. The reason why I consider this relationship to involve the 
second member of the prosodic opposition is that in certain South Welsh 
dialects one does find a 'voiced' variant ([z]) which in some cases is 
contrast ive (see, for example C, Thomas 1964) and suggests a set of 
relationships patterned after those found in the secondary obstructions. 

Unlike the nasal member of the opposition of nasality, aspiration 
can occur without entering into a relationship with a member of the 
obstruction opposition. In this event, the opposition takes on the 
nature of a privative opposition (either there is preaspiration or there 
is not), and the aspirate member of this opposition assumes the charac- 
teristics of the syllable transition on which it is superimposed. 

Not only do the prosodies relate with the obstructions, but they 
also relate with each other. In fact, such a relationship between the 
prosodies is obligatory in Welsh if the nasal member of the opposition 
of nasality is to be realized. In saying that the nasal member of the 
opposition of nasality can only be realized in relationship to the 
primary obstructions of the 1st degree, we are actually saying that the 
nasal opposition member can only occur in a relationship with the second 
and third members of the opposition of aspiration, for it is the occurrence 
of the opposition of aspiration which serves to differentiate obstructions 


of the 1st and 2nd degree in Welsh. 

Due to this dependency relationship between the opposition of 
nasality and that of aspiration, we can base the restrictions placed 
upon the opposition of nasality at least partially upon the aspirate 
prosody. We can thus restate the above-mentioned restriction as follows » 
The nasal member of the opposition of nasality is restricted to the 
primary obstructions constrained by the second or third member of the 
opposition of aspiration. If, for some reason, a primary obstruction 
were to be constrained simultaneously by the nasal member of the 
opposition of nasality and by the first (weakest) member of the opposi- 
tion of aspiration, the nasal member simply could not be realized. 
This is an important consideration in the alternation system, as we see 
in the next section. Once again, this only applies to Modern Welsh, 
not to Old Welsh, 

On account of this restriction placed upon the prosodic opposi- 
tion of nasality, we might wish to make a further partition in the 
obstruction division. This restriction would enable us to consider the 
nasal prosody to be a constraint upon the aspirate prosody as well as 
upon the primary obstruction. Thus, in our description of Welsh phonology, 
we could place nasality above aspiration (consistent with the hierarchical 
model in Table 8,3). Of course, such a device would be a convenience 
for conceptualization, while the notion of constraint would be a state- 
ment of the structure of the phonology.-' 

Now that we have defined the relationships between the various 
members of the oppositions in the obstruction division, we can construct 
a grammar representing the consonant subsystem of Welsh. This grammar 
represents those members found in Standard Welsh and can be conceptualized 


in Table 9.1. a. In this table, we find the members of the various 
oppositions in the hierarchical order consistent with the hierarchical 
model and the data (Table 8.3). 

Table 9.1. a gives the members of the oppositions, but not the 
relationships into which these members enter. These relationships are 
shown in Table 9.1.b. This table can be read as follows: The nasal 
member of the opposition of nasality may constrain (enter into an 
hierarchical relationship with) the second or third member of the 
opposition of aspiration and (simultaneously) with any primary obstruction. 
Any member of the opposition of aspiration may constrain any primary 
obstruction. The first and second member of the opposition of aspira- 
tion may constrain any secondary obstruction. The second member of the 
opposition of aspiration must constrain the tertiary obstruction. 

As far as the individual realizations of these relationships is 
concerned, we need only refer to Table 7.2.b, above. The only change 
which we might want to make in this table is to shift all notations so 
that they conform to the vertical arrangement of Table 9.1. a. Of course, 
such a change is only a convenience for conceptualization, and it would 
have no real bearing on our analyses. 

One point that needs to be prominently stressed is that the notation 
is not the notion. Although such a point may be readily understood, 
linguists in the past have often confused the two, particularly when 
these linguists have been working in schools with highly-developed 
notational systems. For example, one occasionally hears a generativist 
saying that stratificational grammar is just a 'notational variant' of 
transformational -generative grammar. Of course, to those who are 
acquainted even in the slightest with the notions of the two schools, 


Table 9.1. a 

The Obstruction Division of Welsh 

(Consonant Subsystem) 



i h 


b d g 

1 3F 



Prosodic Oppositions 

I Obstruction Oppositions 

Table 9.1.b 
Welsh Consonant Relationships 



u. h hh 

Primary Obstructions Secondary Obstructions Tertiary Obstruction 


such a statement is absurd. On the other hand, to the linguist who knows 
the notions of one school and sees similar notational devices in the 
other, the confusion between the two schools can be quite convincing, 
(Compare Anttila 1972iSection 17.1.) I stress this point, because I 
use notational devices traditionally associated with several different 
schools, and the use of these notational devices here should in no way 
be construed as isomorphic with the use of these devices in other 
schools. For example, the notation /g/ used herein only serves to 
designate the velar member of the opposition of position of obstruction 
and does not represent closure or voice, as the same notation does in 
some other schools. 

Nor should any juxtaposition between any notational devices be 
misconstrued as representing a process relationship or derivation. 
Although I would not rule out any use of process at this point in the 
development, I would not require that relationships be based upon this 
notion. Such a requirement would thwart the flexibility of the system 
(compare Chapter 6 and section 9.0). 

One aspect of the system which should be apparent from the notions 
presented in this section is the high degree of interdependence between 
and independence of the various opposition members in the obstruction 
division (indeed, this should be the case throughout the entire grammar). 
As far as the interdependence of the opposition members is concerned, 
no constraint upon the vocalic pattern can be realized without the 
realization of one member of the prosed ic opposition of aspiration. 
Thus, the obstruction opposition member is dependent upon the prosodic 
constraint. Moreover, it is further dependent upon this constraint for 
the realization of the degree of obstruction (1st, 2nd, or 3rd), 


As far as the independence within the system is concerned, we find 
a high degree of variability within the classifications of obstruction 
opposition members. By labeling the classifications in Table 9.1.b 
rather than listing the members, we effectively state that these rela- 
tionships will hold regardless of the actual member realized. 

This last point of variability can be seen in the development of 
the Welsh affricates (see Griff en 19?^b). As a primary obstruction is 
one in which closure can be realized, the English affricates borrowed 
into Welsh would naturally fall into this classification. The fact, 
then, that in some dialects [c] alternates with |[3j and the fact that 

in one dialect [c] also alternates with [nh] and [j] with [n] can be 


adequately explained and even predicted through an hierarchical grammar 
of Welsh. The new primary obstruction /j/ is merely entering into 
established relationships. Indeed, if this development did not take 
place, then the borrowing of the affricates could be considered anomalous. 

The relationships explained in this section should adequately 
describe the consonant subsystem of Welsh. The question now facing us 
is whether this system of relationships can describe alternations and 
historical change, particularly those described in the previous chapters. 
After all, it would hardly be considered proper to criticize one model, 
only to replace it with another model with the same faults (although, to 
be sure, that would not be without precedent). 

9.2 Mutation, Lenltion, and Provection . The relationships shown in 
the previous section for the consonant subsystem of Welsh provide us 
with a structure through which to consider synchronic alternations and 
historical changes. I should first like to address the system of 


alternations known as the mutation system (see Chapter l). 

On the first, phonetic, level of abstraction, we abstract opposi- 
tions from the basis of characteristics found in physiological and 
acoustic evidence. This abstraction is on a level of phonetic pertinence, 
on which these oppositions serve a phonetic function (representing 
the transmission of communicative sound). On the second, phonological, 
level of abstraction, we abstract oppositions from the phonetic level 
which serve a contrastive function in communication. In the third, 
morpophonological level of abstraction, we abstract oppositions from 
the phonological level of abstraction which serve a grammatical function 
in communication. 

Unlike the lower two levels of abstraction, however, the morpho- 
phonological level of abstraction, at least at this point in the develop- 
ment of the phonology, does not have a structure such as that found 
in Table 8.3. Moreover, there are no notations for entities of morpho- 
phonological pertinence. The reason for this seemingly underdeveloped 
state of morphophonology is that there are no entities, no members of 
oppositions, on the level of morphophonological pertinence. A 'morpho- 
phoneme 1 in this approach is the statement of a relationship that obtains 
between members of an opposition of phonological pertinence, a relation- 
ship that reflects a grammatical function (while contrast serves a 
communicative or semantic function). Thus, the level of morphophonological 
pertinence is in one respect more abstract than and in another respect 
equally as abstract as the level of phonological pertinence.-* 

The Welsh mutation system is a morphophonological system in that 
it is a system of relationships between various members of the oppositions 
of aspiration and nasality reflecting certain grammatical relationships. 


Before we can investigate the actual morphophonemes of Welsh mutations, 
we must first establish the grammatical relationships. These relation- 
ships are precisely those described in Chapter 2. In the hierarchical 
approach, however, we address the environments of Chapter 2 through a 
system of markedness (in the Prague sense). 

The application of markedness to the grammatical environments of 
the mutation system is really quite simple. If a grammatical context 
corresponds to the realization of soft mutation, we can say that the 
context is marked for soft mutation. If it corresponds to the realiza- 
tion of nasal mutation, it is marked for nasal mutation. If it corresponds 
to the realization of spirant mutation, it is marked for spirant mutation. 
And if it corresponds to the realization of aspirate mutation, it is 
marked for aspirate mutation. Now if the grammatical context does not 
correspond to the realization of any mutation, we can say that the 
context is unmarked. 

The application of this grammatical markedness to the morphophoneme 
is somewhat more complex. If the grammatical context is unmarked for 
mutation, then there is no morphophoneme, no relationship at this level 
between one member of either the opposition of aspiration or the 
opposition of nasality and another member of the same opposition. If, 
on the other hand, the grammatical context is marked for, say, soft 
mutation, then there is a morphophoneme which represents a relationship 
between the member of the opposition of aspiration or nasality which would 
be realized if the grammatical context were unmarked and that member of 
the opposition of aspiration or nasality which is in fact realized. 

The difference between the hierarchical approach and the generative 
approach of the first five chapters is significant. First of all, the 


morphophoneme of hierarchical phonology is not an item, but a relation- 
ship? while, on the other hand, the underlying segment of generative 
phonology is an item. The system of markedness in the grammatical 
context relates to the morphophonological level of pertinence in the 
hierarchical approach, but in the generative it is the grammatical 
environment which motivates a derivation from the underlying segment. 
Indeed, the entire structure of the hierarchical phonology is established 
upon the basis of relationships coexisting in a system, not upon the 
generative notions of cause-and-ef f ect , directional derivation from 
one item to another. 

The absence of items in the morphophonological relationships of 
hierarchical phonology must be stressed. In abstraction theory (see, 
for example, Korzybski 1933 'Chapter 25), we cannot abstract an item — only 
a quality of the item or notion about it. The items of phonology exist 
only in the actual acoustic and physiological events of speech (and, 
of course, in the neural connections employed in speech). Thus, we 
abstract oppositions from characteristics found in acoustic and physi- 
ological measurements, measurements which are abstracted from the events 
themselves. As mentioned in section 8,0, the process of abstraction is 
reliable so long as we avoid adding assumptions to the system in the 
process of abstracting. 

The abstraction theory used here differs sharply from the approach 
found in the generative feature system. In the generative approach 
to phonology, the feature is an item found on the underlying level, 
and the value of the item may be changed or replaced through a series 
of processes until it becomes the phonetic event (Chomsky and Halle 
1968i294). In this approach, then, the phonetic feature is just as 


concrete an item as the acoustic-physiological event itself. Unfor- 
tunately, the underlying segment is an abstraction of the distinctive 
features, making this underlying level a level of abstraction. How 
an actual item can be abstracted onto a level of abstraction has never 
been explained, especially as regards unrealized 'abstract' segments. 

In hierarchical phonology, soft mutation can be described as a 
morphophonological system relating the unmarked opposition members to 
their marked counterparts in the opposition of aspiration, the marking 
here corresponding to grammatical context. If in the unmarked context, 
a primary or secondary obstruction is constrained by the second or 
third member of the opposition of aspiration without being constrained 
by the nasal member of the opposition of nasality (except in the case of 
the labial obstruction constrained by the second member of the opposition 
of aspiration), then in the context marked by soft mutation, the obstruc- 
tion is constrained by the next lower member of the opposition of 
aspiration. Thus, for example /bh/ alternates with /b0/, or [p] with 
[b] in the traditional notation. 

These relationships must adhere to the consonant relationships 
outlined in the previous chapter. When we say, then, that the soft 
mutation morphophonemes relate the unmarked member with the next lower 
member of the opposition of aspiration, we must take into acount 
restrictions placed on consonant relationships. Thus, as the /m/ member 
of the opposition of aspiration constrains the /g/ member of the obstruc- 
tion opposition such that the phonetic level realizes null, the realiza- 
tion of soft mutation in the morphophoneme relating unmarked /g0/ 
with marked /gV relates to the realization of null on the phonetic 
level. Likewise, as the nasal member of the opposition of nasality 


cannot be realized with the /m/ member of the opposition of aspiration, 
the realization of soft mutation in the morphophoneme relating unmarked 
/hn/ (or /b0n/) with marked */V\nf is actually realized /bM/ and is 
realized as such on the level of phonological pertinence. 

By maintaining the notion that morphophonological relationships 
must adhere to the structure of the phonology, we avoid one of the 
major problems that has plagued generative phonology, particularly with 
relation to the soft mutation rules of section 1.1. The generative 
systematic phonemic rule operates through processes. When the processes 
would create a segment which does not exist, the rule must be constrained 
or another rule must be added. With the hierarchical morphophoneme 
being firmly established upon the structure of the phonologically 
pertinent level, we can establish relationships on the basis of what 
is actually permissible, thus avoiding the rather awkward false steps 
of Zwicky 197^. Moreover, this approach also renders non-existent the 
problem of rule ordering, as no external devices are needed in order to 
relate one part of the structure with another. 

Nasal mutation is structurally the same as soft mutation. In this 
type of mutation, if in the unmarked grammatical context a primary 
obstruction would be realized constrained by the second or third member 
of the opposition of aspiration, then in the context marked by nasal 
mutation the obstruction is further constrained by the nasal member of 
the opposition of nasality. Thus, for example, /bh/ alternates with 
/bhn/, or [p] with [mh] in the traditional notation. 

Likewise, in spirant mutation, if in the unmarked grammatical 
context a primary obstruction would be realized constrained by the 
third member of the opposition of aspiration but not constrained by 


the nasal member of the opposition of nasality, then in the context 
marked by spirant mutation the obstruction is constrained by the next 
higher member of the opposition of aspiration. Thus, for example, 
/bh/ alternates with /bhh/, or [p] with [f] in the traditional notation. 

In aspirate mutation, the situation is a little different, though 
the structural application of the morphophoneme is the same. If in 
the unmarked grammatical context no obstruction would be realized 
constraining the initial syllable transition, then in the context 
marked by aspirate mutation the transition is constrained by the aspirate 
member of the privative opposition of aspiration. Thus, null alternates 
with /h/. 

Thus, we see that all of the Welsh mutations may be described 
through the use of hierarchical morphophonemes relating members of 
prosodic oppositions. Because of the construction of the phonology along 
obstruction-prosodic lines, no more than one degree of any one prosodic 
opposition is involved in any morphophoneme, nor is more than one 
prosody involved in any one morphophoneme. This system, then, is more 
regular than the generative system of process relationships criticized 
in section 6,1, in which conflicting processes create one major class 
from another. 

Moreover, the hierarchical phonology does not rely upon mechanical 
devices to insure the establishment of relationships. Thus, in the 
event that unforeseen sets of relationships need to be established in 
order to make generalizations about the structure of the phonology, 
there should be no external devices prohibiting these relationships. 

There is another type of mutation, traditionally known as 'hard 
mutation'. As pointed out in sections 5.1 and 6.2, this traditional 


term is used to refer to archiphonemic neutralization in word-final 
position, in which 'stops' lose the contrastive function in the opposi- 
tion of aspiration, and the second member of the opposition of aspira- 
tion is realized for all primary obstructions of the 1st degree. 

This archiphoneme is like a morphophoneme in that it is not an 
entity but a relationship holding between members of an opposition. 
The archiphoneme, however, differs in an important way from the morpho- 
phoneme. While the morphophoneme is dependent upon a notion of gram- 
matical function corresponding to phonological realization, the archi- 
phoneme is dependent upon a notion of contrastive function corresponding 
to phonological realization. This functional difference between the 
two types of relationship should not be treated lightly, for it has its 
roots in the very difference between grammar and meaning (or semantics). 

It is with respect to the archiphoneme that we find the notion 
of phonological marking most evident. Phonological marking is found in 
contrastive oppositions and may be compared with the Prague notion found 
in Trubetzkoy 1969. As each individual language maintains its own 
phonological structure, or grammar, markedness depends in the final 
analysis upon the particular language. In Welsh, markedness within 
the gradual opposition of aspiration is gradual, such that between any 
two members of the opposition, the weaker is the unmarked and the 
stronger is the marked. In word-final position, then, we can say that 
there is a position of neutralization such that if a primary obstruction 
of the 1st degree is to be realized, it can only be realized with the 
weaker, second member of the opposition of aspiration. 

In this system, we avoid the many problems encountered by generative 
phonology with respect to archiphonemic neutralization. By recognizing 


a contrastive function as well as a grammatical function, we can 
differentiate between relationships with phonological correlates and 
those with grammatical correlates (compare Griff en 1974a), More 
importantly, by establishing a phonological structure with phonologically 
and phonetically pertinent levels of abstraction, we can make generaliza- 
tions about the way in which the language operates without having to 
relate this language with a preconceived set of universal markedness 
notions which, in reality, prohibitively conflict with Welsh. 

Traditionally related to the notion of 'hard mutation' is provection. 
Now the aspirate and geminate provection discussed in section 7,2 can 
be described through the use of the opposition of aspiration, as 
demonstrated in that section. Here we may find it convenient to describe 
provection through a process relationship. If aspiration spreads its 
dominance from the initial transition of one syllable to the final 
transition of the preceding syllable and if, in addition, a primary 
obstruction constrained by the second member of the opposition of 
aspiration is constraining the final transition of this preceding syl- 
lable, then the aspiration merges with the second member of the opposi- 
tion to create the third member of the opposition of aspiration con- 
straining the obstruction (for example, the realization of el mab hi 

as[i mha: pi] 'her son' in section 7.2). Likewise, if two identical 

primary obstruction members both constrained by the second member of the 
opposition of aspiration should occupy the same position in the syllable, 
then the obstructions should merge and the two second members of the 
opposition of aspiration should combine, creating a single third member 
(for example, tebyg gennyf as [tebik kin i] *I suppose' in section 7.2). 
Where the range of dominance of a prosody is concerned, process relation- 


ships can be useful, but we ought not to limit ourselves to such process 
relationships. Indeed, any particular type of relationship must be 
considered tentative. Of course, these processes are not derivations 
from deep to surface structures. 

Thus, the synchronic alternations of Welsh can be handled through 
an hierarchical phonology. Such a description, moreover, affords a 
considerable degree of flexibility and avoids the pitfalls mentioned 
in Chapter 6. The question remains, however, as to whether it can handle 
the phenomena of historical change. I examine here the two most common 
types of historical change in Welsh, lenition and provection. 

As discussed in section 6.0, the nature of historical change ought 
not to be confused with the method of describing it. The actual 
•change 1 (if we can call it that) is a system of correspondences between 
one generation of speakers and the next. As it does with synchronic 
alternations, the hierarchical phonology relies upon the structure of 
the phonologies of both generations involved to describe the change. 

An historical change can be described in the hierarchical approach 
through a relationship structurally similar to the morphophoneme and 
archiphoneme. The change serves to relate a member of a particular 
opposition used in one generation to the member of the opposition 
corresponding to it used in a subsequent generation. Although there 
is a structural resemblance between the synchronic alternations on the 
one hand and the historical changes on the other, the functional dif- 
ference between the two is of over-riding importance. 

As mentioned above, the morphophoneme is a relationship between 
one opposition member and another based upon a grammatical function, 
and an archiphoneme is a relationship between one opposition member 


and another based upon a contrastive function. The historical change is 
such a relationship based upon an historical function. As the use of 
one opposition member or another in the morphophoneme serves to convey 
grammatical information and the use of one opposition member in the 
archiphoneme serves to convey information about contrast and markedness, 
the use of one opposition member or another in the historical change 
serves to convey information about the particular structure employed, 
whether it be of one historical period or another. 

Historical lenition, then, is an historical change which relates 
a particular obstruction constrained by one degree of aspiration used 
in one generation to a corresponding obstruction of a subsequent genera- 
tion constrained by the next lower degree of aspiration. Likewise, 
historical provection is an historical change which relates a particular 
obstruction constrained by one degree of aspiration used in one genera- 
tion to a corresponding obstruction of a subsequent generation constrained 
by the next higher degree of aspiration. 

By using a relationship with a function of time, this approach 
more closely parallels the actual event of an historical change. More- 
over, by differentiating between functions of grammar and of contrast 
on the one hand and functions of time (generations) on the other, we 
manage to draw a meaningful distinction between the synchronic and the 
historical . 

Furthermore, the hierarchical approach provides us with a set of 
relationships between the synchronic and the historical such that direct 
comparisons can be made between the structures of the phonologies of 
different generations. This set of relationships can function as a 
panchronic statement of the language after the fashion of Welch 1975. 


The general type of relationship used here to demonstrate the 
historical change as a function of time can also be employed in other 
areas of linguistics. For example, we can form relationships between 
structures of different dialects and different styles within the 
same dialect. 

The fact that these different types of relationship are structurally 
similar can lead to greater insights insofar as these different types 
relate one to another. For example, in section 1.0, I point out that 
phonological change during the transition between Brythonic and Old 
Welsh gave rise to the morphophonological system of mutations. As both 
the change and the morphophoneme can be described through the same 
structural relationships, this hierarchical framework affords us greater 
insight into just how such a situation may come to pass. The notions of 
functionalism found in the hierarchical approach insure that we can 
receive the benefits of structural similarity with the safeguard 
of functional difference. 

Finally, with respect to historical change, I should like to address 
the notion of teleological principles. The hierarchical structure, 
being quite similar to the Prague approach, provides us with a system 
of relationships which, as we see with the affricates in the previous 
section, tends toward maximal systematicity, or symmetry. Such a 
tendency provides us further with the basis for a teleological principle, 
such as that found in Jakobson(l 962 il-2, 2-116,202-20). Through the 
avoidance of too extensive a formalism, we further maintain the flexibility 
needed to use such teleological principles. 


9.3 The Deviation . In the previous section, we examine the 'regular' 
alternations and historical changes and find that an hierarchical approach 
to phonology can insightfully describe such situations. In the deviation, 
as discussed in Chapters 3 and 4, however, we find alternations and 
historical changes which are not quite 'regular', at least not as they 
are described in those two chapters. An adequate description of Welsh, 
then, would have to account for the deviation. 

As far as the exceptions and counter-examples of section 3.0 are 
concerned, the generative practice of marking the lexical items is quite 
similar to the hierarchical approach to thisaort of problem. Rather 
than marking a lexical item alone, however, we can make a grammatical 
context unmarked given particular elements in the context. By making 
the context unmarked, rather than marked, we eliminate any use of a 
morphophoneme in the context. Although the two methods of accounting 
for exceptions and counter-examples appear to be structurally similar, 
the notions underlying the difference in approach to marking is fairly 
significant. Whereas the generative will specify a rule and then act 
to restrict the rule or add another, the hierarchical, like the Prague 
and stratificational approaches, will accomplish the task by specifying 
and using relationships that exist within the structure only. This 
practice avoids the overgeneralization-and-restriction tactics which 
lead to the awkward false steps. 

The difference between the generative and the hierarchical approaches 
to the deviation is somewhat analogous to the difference between the 
approaches to exceptions and counter-examples. First of all, in the 
generative approach to the phonology of Welsh, the deviations holds an 
important position. By using the process framework, the generative 


phonologist can compare the various surface realizations and 'recon- 
struct' an underlying voiceless aspirated stop, as we see in Chapter 3. 
Because the voiceless aspirated stop is not realized in the unmarked 
environment (environment I), we must use a minor rule deriving the 
voiced unaspirated stop from the underlying voiceless aspirated stop. 
Synchronically, this minor rule must be applied every time the item 
bearing the deviation is to be generated in environment I, 

It is in the area of historical change that the deviation is most 
important within the generative framework, as pointed out in Chapter k. 
The addition of the minor rule has been a gradual process in the sense 
that the inclusion of the rule in individual grammars has occurred 
gradually over several centuries and the rule has generalized gradually 
to include all points of articulation. The historical change has not 
affected the underlying segment, though, but has only added a rule to 
the grammar. 

This approach to the deviation reflects the total reliance upon 
the process rule. The only way the generative phonologist has to relate 
the various mutation forms is through derivation from a common underlying 
form. Thus, although a change has indeed occurred historically, the 
phonologist must repeat the historical change every time the item is 
to be derived. If this process is not repeated for every derivation 
after the change (cases of restructuring aside), then the entire struc- 
ture of the phonology as it is established in a process framework 
will collapse. 

On the other hand, the hierarchical phonology is not totally 
dependent upon process relationships. This independence, moreover, 
is reflected in the fact that the phonological structure is far more 


concrete than that of the generative model in general, as well as with 
respect to the deviation. 

In the hierarchical approach to the phenomenon of the deviation 
in synchronic terms, we should recognize an important similarity between 
the various words that reflect the phenomenon i They are all function 
words. This fact is important because function words are dependent upon 
the grammatical contexts in which they appear to a far greater degree 
than are content words. For example, when one encounters the word 
[gan] in one context and [a xan] in another, one should note that the 
meaning and the grammatical function of each word is determined by the 
context. Thus, one is liable to learn these function words as separate 
entities in separate contexts. On the other hand, when one hears the 
words [pobl] and [bobl], content words with the meaning 'people*, one 
is liable to consider them as variants of the same form, because the 
meaning is constant — unaffected by context. 

This attitude toward the difference between function words and 
content words is strongly supported by the evidence presented in 
Chapters 3, 4, and 5. When [gan] replaces [kan] as the radical form, 
no change is found in the mutations. When [bobl] replaces [pool] as 
the radical, however [vobl] replaces [bobl] as the soft mutation form, 
I should contend, then, that there are no mutations of certain function 
words productively employed in Welsh — each word is learned in its 
appropriate context. 

In an hierarchical approach, then, what we find in the synchronic 
manifestation of the deviation is really no deviation at all. Morpho- 
phonemes do not apply to function words. We can tell a function word 
by its particular function in the structure of the language. Both 


aspects of hierarchical phonology which make this analysis possible — the 
flexibility of the relationships and the ability to distinguish between 
productive and nonproductive forms — are impossible in a generative 
approach, for the generative model must rely upon process rules to derive 
variants from the underlying form of a 'single word' and cannot distinguish 
functional differences in its explicit formalism. 

As we turn to the historical aspect of the deviation, we find 
simply a classical case of historical change, as it is described in 
the previous section. Taking the change from Old Welsh [kan] to Modem 
Welsh [gan], for example, we find an historical change relating the 
third member of the opposition of aspiration constraining the velar 
member of the obstruction opposition in the speech of one generation to 
the second member of the opposition of aspiration constraining the velar 
member of the obstruction opposition in a subsequent generation. The 
generalization of this change, as described in section 4.1, can be 
accounted for in our teleological principle stating that the structure 
tends toward maximal systematicity (and economy —compare Martinet 1955). 

A significant advantage of the hierarchical approach over the 
generative lies in the fact that the hierarchical description describes 
a change once, at the time that the change is in fact effected. There 
is no need to repeat the change every time the word is to be realized 
in every subsequent generation. 

The deviation is the point of entry in the generative description 
for the notion of a Welsh consonant shift (compare section 5.0). It 
is this problem, after all, which gives rise to the minor rule. The 
question cn&t to be asked as to whether or not the change viewed through 
an hierarchical phonology is of such overriding importance. 


That there is an historical change involved, there is no doubt. 
We see this in Chapter 4. The change consists of the correspondence 
between an earlier primary obstruction constrained by the third member 
of the opposition of aspiration and a later one constrained by the 
second member. This is quite in keeping with markedness notions and 
is not at all 'irregular'. 

It is doubtful that this type of change represents some sort of 
tension within the phonology pressuring other instances of the third 
member of the opposition of aspiration to relate through a temporal 
function to instances of the second member. First of all, when we cease 
to view the language through the process rules of generative grammar, 
we lose the 'evidence' that the process of the change is constantly in 
use in the synchronic grammar. 

To be sure, the change has been productive, in that it has indeed 
generalized to the entire class represented by the primary obstructions 
and has generalized to a large class of function words. But the produc- 
tivity of this type of change ought not to be overrated. The tendency 
to relate the third and the second member of the opposition of aspira- 
tion temporally may not affect all words and may very well be due to 
some further influence. 

For example, the prepositions are usually in a nontonic position 
in the vocalic pattern. The weakening of the opposition may be a 
function of the position in the stress pattern (vocalic pattern prosody), 
not a function of the position in the word. We see an example of this 
in the phenomena involving lenis initials in Welsh borrowings. As I 
point out in Griff en In press a, the change from the weakest member of 
the opposition of aspiration (tension) to the second member is a 


function of the position of occurrence — and this occurrence is in 
content words and tonic position. 

A much more pointed example with respect to the change in the 
deviation is found in sentence adverbials. As noted in section 2.2.aa, 
all sentence adverbials are considered in the environment of soft 
mutation. Thus, we find the adverb gynt [gint] 'formerly' and ynghynt 
[enhintl 'quicker, earlier'. Of course, to say that soft mutation 
has applied rather than some instance of historical change sometime in 
the past cannot be substantiated, for one never finds *[kint] used as 
an adverb. Thus, we can say that adverbs have undergone an historical 
change (sometime before the coining of ynghynt ). And we can say the 
same with respect to those other function words which exemplify the 
'deviation'. Such an analysis makes no claims that the behavior of 
gan and the others represents a sweeping change in word-initial position. 
In the hierarchical approach, then, the deviation represents a change of 
no greater importance than any other change. 

9.^ The New Welsh Consonant Shift . A pattern of change, when analyzed 
through a framework that does not (and, indeed, cannot) take into 
consideration the notion that structures correspond to functions, 
appears to be indicative of complex relationships and continuous 
pressure for change in the system. On the other hand, when the same 
pattern of change is analyzed through a framework that accommodates the 
notion that structures correspond to functions, that language is both 
systematic and communicative, then this pattern of change appears to be 
little more than routine. 

The loss of special status in the deviation due to the consideration 


of more flexible bases of relationships and of various aspects of the 
communicative function of language raises a question as to the validity 
of the claim that a consonant shift is in progress in the Welsh language. 
As pointed out in section 6.0, the proof of the shift in the generative 
framework does not suffice for a proof of the shift in the language, 
unless we can demonstrate that the generative model is sufficient for the 
description of the language. 

As shown in Chapter 8, the hierarchical model maintains a closer 
relationship with the actual phonetic data. This is accomplished by 
strict controls on the abstraction of oppositions, abstraction of the 
hierarchical order recently determined by phoneticians to be a charac- 
teristic of speech, and the avoidance of assumptions in the process of 
abstraction and construction of the phonological structure. In this 
chapter, moreover, we see that the hierarchical model can be used to 
describe the synchronic alternations and historical changes in Welsh in 
a way which eliminates the points of weakness found in the generative 
description. Thus, inferences based upon this model should be far more 
reliable than inferences based upon the generative. 

In Chapter 5$ the notion that there is a consonant shift in progress 
in Welsh is totally dependent upon inferences based upon the generative 
model. Now that we have eliminated the weaknesses in description which 
result in the weakness in the inference, we must ask whether the 
Inference, like that of the deviation, is capable of surviving without 
the notational devices and the extra-linguistic assumptions. In order 
to determine the reliability of this inference, we must reexamine the 
findings of Chapter 5 in the hierarchical model. 

In section 5.1 i we find a 'word-final voicing' phenomenon which is 


taken to represent the spreading of the minor rule. This phenomenon, 
however, is also described in that section and in section 6.2 as the 
realization of the unmarked opposition member in a position of archi- 
phonemic neutralization. As we see in the data of Pilch 1958 and in 
section 7.2, the phenomenon in question is not 'word-final voicing' or 
even the neutralization of the opposition of voice, but the neutraliza- 
tion of the opposition of aspiration. Voice in such a language as Welsh 
is of secondary, phonetic, pertinence (compare Trubetzkoy 1969«76-7). 

The phenomenon found in word-final position is changed in the 
hierarchical approach, but the hypothesis is in no way weakened. In 
fact, it is strengthened, if anything. In an hierarchical approach, 
we find an archiphoneme (see section 9.2) relating the weak second member 
of the opposition of aspiration to the strong third member. As is the 
nature of the archiphoneme, the contrast ive function obtaining between 
the second and third opposition members is 'neutralized', or lost, and 
the second member of the opposition is realized in the position of 
neutralization. Thus, especially in continuous speech (compare 
Trubetzkoy 1969»262 and section 7.2, above), the degree of aspiration 
is lost and 'voice' is in most cases realized. What is occurring, then, 
in the hierarchical description is not a loss of voicelessness with 
many Modern Welsh exceptions, but a loss of one degree of aspiration 
without exception in Modern Welsh (I return to this below) 

By considering the function of the phenomenon of neutralization in 
word-final position, then, we may very well see ever-more clearly the 
weaknesses of the generative model, but we also see that, given the 
reliability of an hierarchical description, there does appear to be 
an important development in the Welsh language. This development is 


the loss of the contrastive function of the second and third members 
of the opposition of aspiration constraining the primary obstructions 
(obstructions of the 1st degree). In traditional terms, there is a loss 
of contrast between 'stop' cognates. Now this is not an unrestricted 
historical change marked by the temporal function in language, but an 
archiphoneme marked by the contrastive function of language. Neverthe- 
less, as we see below, this loss of contrast has important implications. 

In section 5.2, we find some cluster and intervocalic phenomena 
which support the notion of the minor rule. Looking first at the 
cluster phenomena, we find in Standard Welsh certain developments in 
just what is permissible and what is not in consonant clusters as far 
as voice and aspiration are concerned. Again, voice is only of phonetic 
pertinence, so we should concentrate on the phenomena as they affect 
aspiration. We find several positions in which the second and third 
members of the opposition of aspiration contrast, and other positions 
in which we find only the second member of the opposition. Moreover, 
from the inherited words and from borrowings we can see that the 
second and third members cannot maintain contrast in these positions — 
only the second member is realized. Once again, we find an archiphoneme, 
precisely on the basis as the one found in word-final position. The 
phonological context may be more complex than in the archiphoneme 
above, but the fact that context, or position (as opposed to grammatical 
function), is the determining factor (or, more precisely, the coexisting 
one) indicates that we indeed have an archiphoneme. 

In the intervocalic data presented, we cannot say that we have 
found an archiphoneme, for an archiphoneme operates with regard to a 
position and not with regard to a particular word or set of relationships 


existing between certain words. What we find here is simply a classical 
case of historical change. For example, in the [teked] > [teged] 
change, we find an historical change relating the third member of the 
opposition of aspiration constraining the velar member of the obstruc- 
tion opposition in one generation (/gh/) to the second member of the 
opposition of aspiration constraining the velar member of the obstruction 
opposition in some subsequent generation (/g$/). Now the fact that the 
uninflected form is [teg") and not *£tek] can form a relationship with 
the historical change — the relationship of analogy, a function of 
systematicity. By introducing the analogy, we further show the weaknesses 
of the generative framework in historical linguistics, for the generative 
model cannot in its explicit formalism refer to analogy, so any change 
must be weighed equally with any other change. In the hierarchical model, 
on the other hand, an historical change related to an analogy tends to 
serve the contrast ive function of speech, not to suspend this function, 
as we find in the archiphoneme. As I show below, such a change, based 
only upon analogy and without suspension of contrast, cannot be used as 
evidence for a consonant shift, at least not in an hierarchical model. 

In section 5.3, we examine the phenomena found in word-initial 
position classified as the gan -type change and the pobl -type change. 
As we see in the previous section, the gan -type change is a case of 
historical change. However, if this change can be connected with the 
particular position in the vocalic pattern, probably a nontonic position 
such as in clitics (though this is purely hypothetical), such a change 
may be related to an archiphoneme. In such an archiphoneme, contrast 
between the second and third members of the opposition of aspiration 
constraining the primary obstructions would be suspended, such that only 


the second member would be realized. Such an archiphoneme, though, 
would not be realized because of the position in the word, but because 
of other considerations (such as position in the vocalic pattern combined 
with notions of function). 

In the content words, we find very little evidence at all for a 
consonant shift. To be sure, there is the pobl -type change, described 
in section 5.3, but this description is made without regard to contrast ive 
function. A closer look at this sort of change shows that this is a 
simple case of historical change, of little consequence to the notion of 
a consonant shift. For one thing, the simple change relating the third 
member of the opposition of aspiration constraining the labial member 
of the obstruction opposition in one generation (/bh)/ to the second 
such member in a subsequent generation (/b0/) is of little importance, 
because this change is isolated. Unlike the generative description, an 
hierarchical description can take this into account through an examina- 
tion of the function of contrast. So long as the phonological structure 
maintains contrast between the second and third members of the opposition 
of aspiration constraining the primary obstruction (of the 1st degree) 
in word- initial position of content words, any pobl -type change may be 
simply a case of random change, for the maintenance of the contrast ive 
function in the phonological structure actually thwarts any sweeping 
changes of the magnitude of a consonant shift. Moreover, the pobl -type 
change does not exhibit any change in the structure of the phonology, 
for the morphophonemes relating to the mutation system remain intact, 
a point of importance as shown below. 

Insofar as the mutation morphophonemes and the phonological structure 
are concerned, I should also mention briefly the changes noted in the 


Ty Ddewi dialect. The middle generation shows a distinct loss of contrast 
between the nasal member of the opposition of nasality constraining the 
third member of the opposition of aspiration and that constraining the 
second member of the opposition, resulting in the loss of the stronger 
member from the structure (for example, /dhn/ > /dn/, or [nh] > [n] 
in the traditional notation). This is indeed a shift, in which all of 
the former nasals become realized as the latter. The subsequent 'loss 
of the nasal mutation' in the younger generation is not a loss of the 
morphophoneme, but the relationship of the morphophoneme to nasality. 
Thus, the corresponding nonnasal members are realized — those which, 
due to the above shift, occur with the second member of the opposition 
of aspiration. The generative use of the minor rule here merely reflects 
the demands of the system of notation and process, in which the former 
•deep' underlying segment must be reconstructed in order for processes 
to apply in an orderly fashion. 

As can be seen in the treatment of the change in the T$ Ddewi 
dialect, the central notion in any hierarchical description of a con- 
sonant shift concerns the maintenance of the contrastive function in 
the phonologically pertinent level of abstraction (the level of contrast). 
In this framework, there are three factors involved in a consonant 
shift. The first factor is the loss of contrast between two members of 
an opposition. Using the existing relationships discussed in the 
previous section, such a loss of contrast could be described through 
the use of an archiphoneme. The extent of the archiphoneme in the various 
positions in the vocalic pattern corresponds to the extent of the shift, 
whether it be a shift with a restricted environment or one with an 
unrestricted environment. 


This factor reflects a 'natural' process in the development of 
a consonant shift in language. In the development of a shift, we find 
one generation with a phonological structure such that two members of 
a given opposition serve a contrastive function — represent two 'sounds'. 
In a subsequent generation, these members of the opposition cannot serve 
the same contrastive function, if we are to say that a shift has occurred, 
for if they did, the structure would remain the same and there would be 
no shift, no change, at all. 

The fact that we cannot have in a shift a simple relationship of 
historical change but need an archiphoneme leads to the second factor 
in the hierarchical description of a consonant shift. There must be a 
systematic restructuring of consonant relationships; that is, the 
structure of the oppositions in the grammar (compare section 9t0) 
must change, and this change must be reflected on the phonologically 
pertinent level of abstraction. By a systematic restructuring, I mean 
that the complete system of phonological relationships of the grammar 
of one generation must correspond to a different complete system of 
phonological relationships of the grammar of a subsequent generation. 

Now this structural requirement corresponds to a functional require- 
ment by the nature of the phonological system upon which it rests. Any 
change in the phonologically pertinent level is going to be reflected 
in functional changes in the system of contrast. Thus, a consonant 
shift can in no way be confused with a 'rephonemicization', a simple 
change of phonetic correlates of phonological relationships without 
any structural change. By using the notions of function and contrast 
together with the notion of phonological structure, an hierarchical 
phonology can make such a distinction. 


The final factor in the description of a consonant shift in the 
hierarchical approach is the presence of the temporal function. This 
is to say that we can only refer to a shift between two generations 
(not necessarily adjacent), if the one generation has the one structure 
and the other generation has the different one. Although the presence 
of an archiphoneme in the structures of a succession of generations may 
imply a restructuring and a shift at an earlier time, we cannot claim 
that a shift has taken place during that particular succession of 
generations, for, insofar as that particular succession of generations 
is concerned, no function of time is served. 

These three factors not only provide us with procedures for 
describing consonant shifts, but they also serve as a teleological 
principle for the prediction of consonant shifts. As we examine the 
system of relationships between the various consonant oppositions in 
Welsh, we find that there is a finite (although, to be sure, a large) 
number of possible oppositions that could be used to describe the 
structure. Changes would occur with respect to these oppositions 
and in keeping with markedness notions, such as those of Trubetzkoy 1969. 

We have reviewed the arguments of Chapter 5 in an hierarchical 
approach and have determined the factors needed to demonstrate a shift 
in this approach. The main question before us has yet to be answered, 
though. Is there a consonant shift in progress in Welsh? To answer 
this question, at least as far as the hierarchical approach will take 
us, we must interpret the evidence by the criteria of the method. 

On the positive side of the question, there has definitely been 
a loss of contrast. We see this in the archiphonemic neutralization 
occurring in word-final position, certain cluster positions, and the 


rather nebulous area of word- Initial position of (nontonic?) function 
words. If this is a shift, however, it cannot be a shift with an 
unrestricted environment (compare section 5.0), for it is not complete. 
Word- initial position of content words, certain other cluster positions, 
and intervocalic position still support the contrastive function between 
the second and third members of the opposition of aspiration constraining 
primary obstructions (of the 1st degree). To be sure, there have been 
some changes, as we see in the pobl- type changes and the analogies, 
but at present these are not archiphonemic in nature, and In describing 
a consonant shift, the notion of archiphonemic neutralization is of 
utmost importance (where the archiphoneme is opposed to the morphophoneme — 
compare also Griff en 197**a). 

We may have here a consonant shift with a restricted environment. 
As I point out in section 5.0» such developments in a language can 
lead to situations as we find in Modern Danish, in which the two members 
of the 1st degree obstructions serve a contrastive function only in 
word- initial position. Welsh could be affected in much the same way, 
with the shift restricted possibly from certain tonic considerations in 
the vocalic pattern. 

There are also some positive and some negative aspects in the 
answer to the second fact. Where the neutralization has occurred, 
there has been a systematic restructuring of the relationships pertaining 
to the second and third members of the opposition of aspiration con- 
straining a primary obstruction (1st degree). Thus, there is a tendency 
toward the development of a different structure without functional 
contrast between these two opposition members. If such a loss of 
contrast were to continue unabated, in keeping with our notions of 


teleology, we could say that a shift would be inevitable. 

Once again, however, the incomplete nature of the neutralization 
provides diverging evidence. So long as contrast is maintained in some 
position in the vocalic pattern, then the entire structure of the 
phonology must remain intact, If the structure of the one generation 
corresponds in relationships to that of a subsequent generation, we 
cannot say that there has been a shift, at least not in an unrestricted 
environment. Of course, to describe a shift in a restricted environment, 
we must leave the structure of the consonant oppositions proper and treat 
their relationships with the vocalic (syllabic) oppositions as well. 

The pobl -type change is merely a manifestation of random historical 
change as it stands now in the development of Welsh. It may be fruit- 
ful, however, to consider what would be involved if the position of 
neutralization should in some future generation include word-initial 
position of content words. On the basis of the pobl -type changes to 
date, the loss? of the functional contrast between the second and third 
member of the opposition of aspiration constraining a primary obstruction 
in a case where there is a morphophoneme would probably lead to the 
effectual loss of the spirant mutation. Indeed, all mutation morpho- 
phonemes from the former radical would be replaced by those of the newer 
radical (constrained by the second member of the opposition of aspiration). 
Thus, the tendency in the system would be to shift all members of the 
opposition of aspiration one position in the structure, while the fourth 
member of the opposition of aspiration would remain. This would in 
effect eliminate the third member from subsequent grammars. Thus, 
we would see a change in several relationships (compare Grimm's Law). 
Such a situation has not yet occurred, but given the neutralizations 


which have so far occurred, it is a possibility. 

Perhaps the one greatest factor militating against such a change 
concerns the temporal function and a situation which is external to 
the Welsh language. Although the tendency is there, in order for the 
shift to be complete, the shift must be reflected in the phonology of 
some generation. Thus, we do not have a shift at this point. Whether 
the shift will develop more completely in the future depends upon the 
future of the relationship between Welsh and English. 

In the past century, the English occupation of Wales has been 
marked by a conscious attempt to Anglicize the country, particularly 
through education (compare W.R. Jones 1966), to a degree far in excess 
of previous centuries. As we see in Griff en 197^-b, the reliance of 
Welsh-speakers upon English in the bilingual situation can indeed result 
in historical changes in the Welsh phonological structure. As English 
maintains a strict contrast between aspirata (which may include tenuis) 
and media, in traditional terms, the effects of bilingualism could well 
be the maintenance of the same contrastive function in Welsh. Of course, 
such a conclusion is still open to doubt, as the study of bilingualism 
is still fairly young and it has only been a quarter of a century since 
the 'coexistent phonemic systems' hypothesis of Pries and Pike 19^9* 
which has been to some degree refuted in the study of Welsh. 

The development of current phenomena into a consonant shift depends 
to a great extent, then, upon the future of the Welsh culture. If the 
process of Anglicization continues with the preeminence of English in 
the bilingual situation, then I should not expect the shift to progress. 
Indeed, according to Professor T. Arwyn Watkins (personal communication), 
if the present process continues, there may very well not be a Welsh 


language "by the end of the next century. On the other hand, if there 
is a reversal in the process of Anglicization and Welsh becomes pre- 
eminent in Wales, then I might expect the shift to progress, at least 
to a status equal with the situation in Modern Banish. According to 
Professor Ceinwen Thomas (personal communication), there is in fact 
some evidence to support a resurgence of Welsh even in the most 
Anglicized areas. Thus, the linguists examining this phenomenon must 
consider the sociological situation as well as the linguistic and must 
correlate the two in any future conclusions. 

In any case, the determination of the progress of the shift must 
rest upon a reliable correspondence between the framework used in 
description and the language itself. Such a correspondence depends 
upon the degree to which the framework uses relationships observed in 
the actual data of the language, as opposed to relationships supplied 
from logic or mathematics. It further depends upon the degree to which 
the framework can successfully avoid adding assumptions not based upon 
observed phonetic evidence to the structure of the phonology. Moreover, 
it depends upon the flexibility of the phonological relationships, to 
insure that it is the linguist's structure which adapts, not the data 
which is made to adapt. Finally, it depends upon the recognition by 
the linguist of the fact that language serves a communicative function, 
manifested in the semantics and grammar, and is not merely a structure 
existing in a vacuum. The hierarchical model meets these criteria. 


l-Some dialects (for example, Bangor — see Fynes Clinton 1913 i 
xxxii-iii) maintain the 'groove fricative* [I], which also falls into 
this category, though it could probably be handled as some type of 
prosodic variant of the obstruction /s/ (as the dialectal [zj is 
handled through the prosody of aspiration — see below). 

2This makes no claims as to the realizations of liquids in any 
other language but Welsh, as each language represents its own par- 
ticular structure, 

3l should emphasize that a conceptual device — placing something 
above something in a table — ought not to be confused with the notion 
of relationships which the device is an attempt to convey. 

TThe term 'mutation' is traditional in the study of Welsh, and 
I retain it here so as to avoid confusing the issue with too much new 
terminology. I should note, however, that the notion of mutation 
traditionally implies a process relationship. For example, in soft 
mutation the radical is traditionally considered to change into the soft 
mutation form through the process of lenition. As shown in Chapter 1, 
this traditional notion is quite in keeping with the generative phono- 
logical approach, in which the underlying form changes into the phono- 
logical segment through some sort of process. 

Although I maintain this term, I do not maintain the traditional 
(or generative) reliance upon the process framework (at least, not 
exclusively). In an hierarchical phonology, the mutation system is 
represented by way of relationships (for a further discussion of 
mutations and relations, see Lamb 1975). These relationships are 
morphophonological , in the Prague sense, necessitating a further 
level of abstraction. 

5Note that while we do refer to a morphophoneme, there is no such 
element in this system as a 'phoneme'. Nor is the morphophoneme a 
thing or entity. One could, by the same token, speak of a phoneme as 
a statement of a relationship between opposition members that serves 
a contrastive function. I shall not press this point, as there is 
already much confusion in the linguistic field from a proliferation 
of various 'phonemes' (compare Lamb 1966a), 

6jioreover, considering each possible change and combination of 
changes, we can construct an as-yet-unrealized structure of some future 
phonology of Welsh. Certainly, it would be a tedious exercise, but it 
is possible nonetheless. On the other hand, such an exercise would be 
impossible in the generative model, for without markedness, changes can 
only be described as isolated changes (in much the same manner as the 



traditional process of the Neo-grammarians — compare Vachek 1966:16), 
and with markedness changes can only be described in relation to a set 
of universal interpretive conventions shown in section 6.2 to be 
unworkable, at best. 

?This is more accurately described as the failure of one generation 
to construct a phonology consistent with the contrast patterns of the 
previous generation. 


10.0 Phonetic Justification . Throughout the previous three chapters, 
I stress the importance of the inner approach to phonology (Jakobson 
and Halle 1962). The notion that the various abstracted oppositions 
and relations should maintain a close correlation with the observed 
phenomena in phonetics is not merely a device used to insure the con- 
sistence of the structure of the phonology. Indeed, if consistence 
were all that is needed in a phonological structure, we could greatly 
simplify our task by taking a treatise on logic, replacing logical 
symbols with letters from the international phonetic alphabet, and 
presenting the result as a logically consistent phonological structure. 
Phonology, however, has other requirements besides internal consistence, 
for it is a part of the complex, symbolic communications system used 
by human beings (compare Yngve 1975) » or, at least, of a description 
thereof. Without a basis in the actual communicative act of speech, 
phonology would lack the necessary reference to the real world essential 
in such a system, and it is through the inner approach that we can 
supply the phonology with the physical as well as with the logical 
aspects of speech, and we need consistence with both in the analysis 
of language (compare Hjelmslev 1961:5-6). 

The inner approach, then, is an indispensable tenet of the 
current theory of language. Because it is basic to the theory, I 
use the inner approach in the construction of hypotheses (themodel) 



for the description of observations (compare section 9.0). The 
question remains, however, as to whether the construction of the 
phonology in the previous three chapters is adequate insofar as the 
representation of the inner approach in the description is concerned. 
In order to test these hypotheses, we must examine the elements of the 
model with respect to the functional/structural theory and the observa- 
tions. The direction of our inquiry is, on both counts, phonetic, 
for only by examining the phonetic evidence can we test the phonology 
with respect to the inner approach and the data upon which that approach 
is ultimately based. Of course, such a test represents the logical 
process of abduction, and the results of such a test, while deter- 
mining validity in the logical sense, ought not to be taken as final. 

This notion that the phonology ought to be tested through the 
phonetics is in keeping with most approaches (if not in practice, then 
in theory), though certainly not with all (compare, for example, 
Foley 1970b), Even in those theories which do recognize this relation- 
ship between phonetics and phonology, however, tests are seldom performed. 
Although experimental phonetics has provided the phonologist with evidence 
for oppositions or features (compare Fant 1973)» there has been very 
little substantial evidence provided by the experimental phoneticians 
for phonological structures, and that evidence which does exist (for 
example, Mermelstein 1973 — see Chapter 8) is seldom regarded by linguists. 
Indeed, phonologists have often gone to great lengths to demonstrate 
the consistence of their phonological structures and have created 
elaborate sets of relationships but have 'postponed* the establishment 
of a phonetic basis until some later time. For example, the 'dependency 
model' of Anderson and Jones 197^ comes very close to suggesting the 


hierarchical model, but, although it does make use of features (though 
these are structurally rather than phonetically based — compare Foley 
1970b), the system has not been devised through the inner approach, and 
the phonetic justifications for the relationships have been left to some 
indefinite time in the future (Professor Charles Jones, University of 
Edinburgh, personal communication). 

In spite of the traditional hesitation on the part of many phono- 
logists to test their systems against real phonetic data, I would 
maintain that if we are to construct a phonology which adequately 
represents the relationship between the symbolic language and the 
phonetic speech, we should spend no less time and effort in the area of 
phonetics (the explicit relationships) than we spend in the realm of 
logic (the implicit relationships). In the construction of the hierar- 
chical model and its application in the phonology of Welsh, then, I 
have based my phonological relationships upon evidence supplied from 
phoneticians. The basic structural relationships, for example, reflect 
the findings of Mermelstein 1973» Cbman 19^7, and others, as I mention 
in Chapters 8 and 9. By maintaining the close correlation with the 
phonetic evidence in the construction of the model, the model is, in 
effect, tested with respect to this phonetic evidence. 

The construction of the model, however, is just one aspect of 
hierarchical phonology. Another very important aspect which must be 
tested is the use of oppositions within the model. These oppositions 
fall into two categories » the obstruction opposition and the prosodies. 
Do the members of these oppositions adequately represent observable 
phenomena in the phonetic data; that is, can they successfully be tested? 

The use of a! opposition of position of articulation 


(obstruction) and a privative opposition of nasality are, as mentioned 
in Chapter 8, well founded in the literature of acoustic and physi- 
ological phonetics. The nature of these two oppositions has been 
tested and retested, and the validity of using the two oppositions in 
their present form in a model of phonology is hardly objectionable, 
at least from the basis of the phonetic justification. There is one 
final opposition, on the other hand, which is quite open to objection, 
for it has no previous justification in the phonetic literature. This 
is the gradual opposition of aspiration. 

The entire phonological system of Welsh, as I present it in the 
previous chapter, hinges upon the nature of the gradual opposition 
of aspiration. It is this opposition which determines whether the 
resulting 'sound' is to be a fricative or a stop, two classes which 
do have acoustic and physiological justification in the literature. 
Because of the importance of this gradual opposition to the phonology 
of Welsh, as it is presented in the previous chapter, and because of 
the absence of justification for this opposition in the literature, I 
present acoustic and physiological justification for it here. In 
addition to the acoustic and physiological justification, there is also 
some behavioral justification for the opposition as it is used in Welsh, 

10.1 Acoustic Basis of Aspiration . One area of phonetics from which 
justification for the phonological use of aspiration must come is the 
area of acoustic phonetics. Although acoustic phonetics is considered 
the only valid route for the inner approach in Jakobson and Halle 1962 
because of its reliance upon the effect rather than the cause of speech 
sound, more recent linguistic treatments of phonology, such as Chomsky 


and Halle 1968, tend to avoid the acoustic justifications of phono- 
logical oppositions and structures in favor of the articulatory. Thus, 
before I use evidence from acoustic phonetics in order to justify my 
use of aspiration in the structure of Welsh, I should first provide a 
brief explanation, giving my reasons for maintaining the more tradi- 
tional and less popular notion that acoustic phonetics is pertinent to 
the study of phonology. 

The analysis of the speech event consists of two phonetic parts — 
the physiological, in which the speaker's vocal apparatus and the 
hearer's auditory apparatus are examined, and the acoustic, in which the 
sounds transmitted and received are examined. To be sure, there should 
be some correspondence between the two (compare Ladefoged 1962:Chapter 7), 
but to study the physiological aspects of speech without considering 
the acoustic would be to assume that the configuration of the speaker's 
vocal apparatus is physiologically transmitted to the hearer's auditory 
apparatus, that each and every aspect of the phsylological state of the 
speaker is transmitted without loss to the hearer and is important for 
communication to the hearer, and that the medium -of transmission neither 
adds to nor detracts from the actual message. Before we can safely rule 
out the acoustic aspects of speech, then, we must first justify these 
assumptions. If, moreover, these assumptions cannot be justified, then 
the acoustic aspects of speech had best not be avoided. 

The first assumption, that the physiological state of the speaker 
is transmitted to the hearer's auditory apparatus through some physi- 
ological means is false, for there is no physiological connection 
between the speaker's mouth and the hearer's ear in normal speech 
communication. The speech sounds, as acoustic entities, are not only 


produced by the speaker, but they are understood by the hearer. This 
latter point, that speech sounds are understood by the hearer in com- 
munication, may appear to be trivial, but linguists have tended to 
ignor it in the light of recent developments. 

The recent developments which have obscured the importance of 
acoustic phonetics in linguistic research have to do with the motor 
theory of speech perception (compare Halle and Stevens 1964), In the 
motor theory, or analysis by synthesis, the auditory stimulus is held 
to be translated into articulatory responses which would be necessary 
if the hearer were to produce the sounds heard. It is further maintained 
in this theory that the perception of speech is not an act of decoding 
the auditory stimulus, but an act of decoding the articulatory response. 
Thus, incidently, it maintains the notion that it is the effect, rather 
than the cause, which should be analyzed. There is considerable evidence 
for this theory, and I do not challenge the theory per se, only the logic 
of using this theory as evidence against the analysis of acoustic 

The translation of the auditory stimulus into an articulatory 
response presupposes the reception and comprehension of the acoustic 
stimulus by the auditory apparatus. An argument, then, against the 
importance of the acoustic analysis on the basis of the motor theory's 
notion that it is the articulatory response rather than the acoustic 
stimulus which serves the phonology violates the logical notion of 
transitivity and is, as such, invalid. It is the acoustic stimulus 
which brings about the articulatory response which, in turn, yields 
the phonological analysis; so that, in the analysis of speech perception, 
the articulatory response and the phonological analysis are both 


dependent upon the acoustic stimulus. Were we to disregard the acoustic 
stimulus, we would have no evidence for the articulatory motor response 
and no analysis. 

Moreover, the final step in the procedure of analysis by synthesis 
rests upon the acoustic signal. As pointed out in Chomsky and Halle 
(I968t2°4-), the synthesized product must be acoustically compared with 
the stimulus in order to insure accuracy. Were we to disregard the 
acoustic stimulus, we could further have no way of verifying the results 
of the analysis — the important final step in the process. The motor 
theory, then, in no way detracts from the necessity of understanding 
the acoustic signal; rather, it underscores it. 

Nor should this argument be construed as an attack upon physi- 
ological phonetics. Both the physiological and the acoustic phonetics 
are necessary for a complete analysis, for each interacts with and is 
dependent upon the other (compare Lieberman 1970). 

In the current acoustic theory of speech perception (see Fant i960), 
the acoustic signal is analyzed into functions of frequency, amplitude, 
and time with respect to the sound waves used in speech. My acoustic 
argument for the gradual opposition of aspiration rests solely upon 
spectrographs analysis in keeping with the current acoustic theory. 
Of course, as mentioned in section 8.1, the theory is not yet theorem, 
but it does reflect the extent of current phonetic knowledge in this area. 

In testing the validity of the opposition of aspiration in the 
acoustic phonetic area, we must also bear in mind that the framework is 
hierarchical, as proposed in Chapter 8 and developed in Chapter 9. 
Once again, we are proving the validity of an opposition as part of an 
overall phonological system by testing the opposition in its proposed 


form for consistence with both the observed data and the theory. 
Although, as mentioned in the previous section, the proof is abductive, 
it is not circular in the sense of using an hypothesis to prove itself. 

A few points should be reiterated, points which are of special 
importance in the hierarchical framework as far as aspiration may be 
concerned. First of all, consonants are constraints on vowels in this 
framework. The members of the obstruction opposition directly constrain 
the syllable transitions, while the obstruction prosodies, such as 
aspiration, further constrain the obstructions. The nature of the 
prosody as a constraint should be provided in the acoustic evidence, 
if our test is to be affirmative. Moreover, as the opposition of aspira- 
tion operates in the phonology as a gradual opposition, we should find 
evidence from the examination of acoustic data for this gradual nature. 

The analysis of the acoustic evidence can best be presented in the 
form of a phonetic experiment. The procedure consists of gathering 
and analyzing the data. The data was gathered in June 197*4- in field 
conditions in Wales, Responses from the four informants mentioned in 
section 5.2 were recorded on a Uher 4000 Report L tape recorder at a 
speed of 7.5 inches per second and a distance of approximately 1.5 feet. 
The following corpus, in standard orthography, was read by each infor- 
mant i 

Orthograp hy Broad Notation 

pahap pahap 

tahat tahat 

cahac kahak 

bahab bahab 

dahad dahad 

gahag gahag 

ffahaff fahaf 

thahath OahaG 

chahach xahax 

fahaf vahav 


Orthography Broad Notation 

ddahadd 6aha6 

llahall lahal 

rhaha saha 

lahal lahal 

rahar rahar 

haha haha 

sahas sahas 

As in normal Welsh words, stress was on the penult and pitch on the 
ultima (compare D.M. Jones 19^9). 

For analysis, the data was rerecorded from an Ampex PR-10 tape 
unit onto the single track Voice Identification 700 also at a speed of 
7.5 inches per second. The utterances were then analyzed on the Voice 
Identification 700 sound spectrograph. The resulting spectrograms were 
organized in accordance with the structure of relationships based upon 
the opposition of aspiration, as found in Table 7.2.b (the nonnasal 
columns), in order to determine what, if any, correlations could be 
found in the spectrograms. 

The results varied between individual speakers and between various 
obstruction opposition members, but within each individual and within 
each obstruction opposition member, the results were uniform in one 
particular relationship. In each opposition member of each speaker, 
the least aspirated member showed a low high-to-low frequency ratio of 
sound energy associated with the obstruction, the second member showed 
a greater high-to-low frequency ratio of sound energy, the third showed 
a greater ratio yet, and the fourth aspirate member showed a high 
high- to-low ratio. 

An example of this high-to-low frequency ratio can be found in 
Figures lO.l.a-d. These represent the dental obstruction opposition 
member constrained by the aspirate prosody in increasing pertinent 
steps in the speech of Dr. Bedwyr L. Jones. In Figure 10.1. a, the 


occurrence of the least aspirate prosody may be connected with the 
fact that the energy associated with the consonantal constraint is 
entirely concentrated at a frequency lower than 1500 cps in final 
position and 1000 cps in initial position. In Figure lO.l.b, the energy 
associated with the consonantal constraint is distributed more evenly, 
with most of the post-release energy concentrated below 3000 cps in 
initial position and with an 'off -glide' in final position, with most 
of the energy below 1000 cps and some below 500 cps. In Figure lO.l.c, 
the energy associated with the consonantal constraint is entirely 
above 500 cps in the post-release stage and with heavier concentrations 
above 3000 cps and traces above 5000 cps in initial position, while in 
final position (which this speaker pronounced in careful speech) the 
energy concentrations above 2000 cps are by far stronger than those 
below 2000 cps. In Figure lO.l.d, the energy associated with the 
consonantal constraint is almost entirely above 1000 cps with traces 
in both positions above 7000 cps. 

As mentioned above, the results vary in detail between speakers 
and between members of the obstuction opposition in individual speakers. 
The variations observed, however, did not affect the basic relationships 
between the prosodic members of each obstruction, as far as the ratio 
is concerned. For example, Mr. Idris Roberts spoke with an affricated 
[t»] (compare section 1.2), reflecting the high-frequency sound of the 
[s] with the release characteristics of the [t] (or /dh/) mentioned 

On the basis of the results, we can conclude that prosodic 
aspiration is a function of the ratio of high-to-low frequency energy 
in the acoustic signal. As such, it is gradual, realized as greater 


* , i 

utterance of [6ahao] 



i i 



Figure lO.l.b 
Utterance of [dahad] 



fi 1 

: 'iK 

Figure lO.l.c 
Utterance of [tahat] 


I - 1 . ii l i m M 






Figure lO.l.d 
Utterance of [9aha9] 


constraining force in greater degrees of aspiration. We should recall 
that in this model, consonants are constraints on vowels and obstruc- 
tion prosodies further constrain the obstruction elements. Since the 
vocalic energy in the spectrogram is concentrated at the lower frequencies, 
as the degree of constraint represented in the degree of the obstruction 
prosody should increase, the obscuring effect upon the vocalic element 
should increase in proportion. This is precisely what occurs in the data. 

Thus, the notion of a gradual opposition of prosodic aspiration is 
not only consistent with the hierarchical phonology, but it is also con- 
sistent with the acoustic observations. These requisites being met, the 
system is consistent and valid, at least so far as the acoustic aspects 
of speech are concerned. 

Before leaving the acoustic basis of aspiration, I should like to 
stress two points. First, the notion of a high-to-low frequency energy 
function for aspiration is hypothesis, as are all analyses. It can be 
taken as fact only within a particular theory, so long as it is consistent 
with the other hypotheses and observations in the theory. As in any 
other science, in linguistics we can only strive for consistence in 
logic, and this approach tentatively satisfies consistence. Second, the 
acoustic relationships demonstrated for the prosody of aspiration hold 
only for Welsh. The phonetic detail supporting the phonological struc- 
ture of one language cannot be generalized to any other language unless 
the other language should maintain the same phonological relationships, 
in which case generalization is still only a possibility. Of course, 
the particular acoustic characteristics of the Welsh 'sounds' will 
probably correspond to the 'sounds' of other languages, just as the loci 
and transitions of obstructions do not very appreciably between languages, 


but the organization using this ratio of aspiration can only be viewed 
as language specific. 

10.2 Physiological Basis of Aspiration . As mentioned in the previous 
section, the act of verbal communication consists of two parts — the 
physiological and the acoustic. In order to give full phonetic justifica- 
tion to an opposition in the phonological structure, we must consider 
both aspects of phonetics. In the previous section, we find that the 
notion of a gradual prosodic opposition of aspiration is successfully 
tested in the acoustic area, that there is evidence in the speech sound 
wave to support the opposition and the use of the opposition in the 
phonological structure. We must now examine the physiological data in 
order to determine what, if any, physiological correlates there may be 
for this opposition. 

We need not, however, begin our search for a physiological cor- 
relate for aspiration without any indication of what to look for. We 
can use the acoustic evidence of the previous section to guide our 
inquiry. The acoustic evidence indicates that this prosodic opposition, 
consistent with the hypotheses of the hierarchical phonology, occurs as 
a constraint with the member of the obstruction opposition. Moreover, 
it occurs with each member of the obstruction opposition in the same 
manner, being realized as a gradual constraint which serves to obscure 
the vocalic element proportional to its own presence. 

Taking into account the constant nature of the opposition of 
aspiration with respect to the member of the obstruction opposition, 
we can probably rule out any significant influence of the articulatory 
musculature used in the production of the various members of the 


obstruction opposition, at least from initial consideration. It is not 
as likely for each set of muscles associated with the different obstruc- 
tions to pattern in precisely the same manner as it is for this constant 
ratio to be caused by the actions of some other part of the physiology 
which affects all obstructions without regard to the particular member. 
The notion that the opposition of aspiration should emanate from a 
physiological characteristic which is not a part of the articulatory 
musculature but which affects all obstructions without regard for 
particular obstructions is an hypothesis of hierarchical phonology (that 
is, it is consistent with the major premise, or theory, involved in 
the test of phonetic hypotheses). Furthermore, this notion also has 
an important precedent—the prosodic opposition of nasality affects 
each of the primary obstructions from a physiological characteristic 
not connected with the articulatory musculature, and this prosodic 
nasality functions in the same manner phonologically as aspiration (as 
a prosody). Thus, we should, at least tentatively, direct our attention 
away from the oral cavity. 

Taking into account the nature of aspiration as a gradual constraint 
upon the vocalic element, we can form a further hypothesis as to the 
location of the physiological characteristic associated with aspiration. 
Acoustically, the greater the realization of aspiration, the more 
obscured is the vocalic element. This vocalic element is basically 
realized through phonation— the vibration of the vocal cords (compare 
Broad 1973). Indeed, the members of the opposition of aspiration can 
be divided into two groups, with the weaker two members insufficient to 
constrain voicing completely and with the stronger two members sufficient 
to constrain voicing completely. Of course, the voicing is an element 


of the vocalic pattern. Thus, in order to locate the source of aspira- 
tion, we should direct our attention to the area around the larynx, 
where constraint of the apparatus used in phonation is possible. Note 
that this direction of inquiry does not follow from — though it is not 
excluded "by — a segmental approach in which aspiration, or tension, is 
inherent, but it does follow from an hierarchical approach. 

On the basis, then, of the evidence from the acoustic data of the 
previous section and the notions of the hierarchical phonology, we 
should look for a physiological characteristic in the vicinity of the 
larynx. This characteristic should reflect the gradual constraining 
force from the weakest 'voiced fricative' to the strongest 'voiceless 

The evidence supporting the gradual prosodic opposition of aspira- 
tion in physiological phonetics is found in the available literature. 
According to Perkell (l969«36-7), the width of the orifice of the 
larynx before consonant release varies in the same proportion as the 
strength of phonological aspiration as it is found in Welsh. Perkell *s 
measurements are taken from a single informant recorded on cineradi- 
ographic film. The precise procedure and results of his experiment can 
be found in Perkell 1969. 

The results as they affect the notion of aspiration constitute a 
test of the theory. In Figure 10.2, I give the measurements in graph 
form for the width of the orifice of the larynx for the nonsense words 
[he'z*], [hs'de], [ha'ta], and [ho'sfc] at the point of widest dispersion, 
approximately 75 msec before consonant release. Phonetically, the [z] 
and the [s] pattern after the fricatives, the weakest and strongest 
members, respectively, of the opposition of aspiration constraining 


the obstruction. The width of the orifice of the larynx for the 
weakest member is approximately 4.1 mm, for the next member 5.6 mm, for 
the next member 7.2 mm, and for the strongest member 9.4 mm. In each 
case, the orifice of the larynx expands beginning at approximately 
150 msec before the consonant release and then contracts to a point of 
minimum dispersion at approximately 20 msec before the consonant release. 



Figure 10.2 
Width of the Orifice of the Larynx 


The cause of the gradual increase in the width of the orifice of 
the larynx is open to some conjecture. Perkell suggests that the 
fluctuation in the width may be due to the forced expansion of the 
walls from some heightened pressure up through the glottis. This 
pressure would meet less resistance during the articulation of the 
•lax' [z] and [d] than it would during the 'tense' [t] and [s]. Thus, 
he suggests that this could be the result of pressure and tension* 
This reliance upon pressure and tension is the mainstay of the traditional 
notion of tension in phonology (compare also Palmer 1964), 


On the other hand, as Perkell also suggests, the widening of the 
orifice of the larynx may be due to an overt muscular gesture. This 
notion is suggested by Ladefoged (1971 « 96-7) with support from Lisker 
and Abramson 1967 and Kent and Moll I969. 

The important aspect of the widening of the orifice of the larynx, 
however, is not what causes it, but what its effect is (compare Jakob- 
son and Halle 1962; section 10.1, above). The effect of the widening 
of the orifice of the larynx is to draw breath into the cavity. This 
breath is then forced out of the cavity as the cavity closes at a 
rapid rate (occurring in approximately 55 msec). The force of the 
escaping breath represents pressure in the articulatory apparatus, 
pressure in a gradual amount. The higher the pressure, the more likely 
it is that the phonation will be obscured and that the puff of breath 
traditionally associated with aspiration will be emitted from the oral 
cavity (or, more precisely, through the oral cavity). 

Thus, the widening of the orifice of the larynx reflects the 
gradual nature of the prosodic opposition of aspiration as well as the 
nature of the opposition as a constraint. Moreover, it closely reflects 
the acoustic situation observed in the previous section. Finally, it 
is consistent not only with the observed data, but also with the 
structure of the hierarchical phonology. As such, then, it is sufficient 
to demonstrate that the system proposed is valid as it stands with the 
gradualopposition of aspiration. As always, we have demonstrated only 
that the system is consistent and valid with the proposed hypotheses, 
not that the system is 'true' in any literal sense. 

10.3 Behavioral Justification of Aspiration . The physiology of the 
vocal apparatus does not vary as a function of the language community. 


Nor do the acoustic characteristics of sound vary as a function of the 
particular language spoken. Thus, we might expect speakers of German 
in the production of some of the 'sounds' of German to realize the 
same physiological configurations and to produce the same acoustic 
characteristics as those produced by speakers of Welsh in the produc- 
tion of some of the 'sounds' of Welsh. Given the species-specific 
physiology and the acoustic characteristics of nature, such a coincidence 
should not be very surprising. 

Although the above-mentioned coincidence may not be very surprising, 
it can be threatening to the phonetic evidence presented in this chapter, 
if we follow either of two extremes. The first extreme would be to 
look at the phonetic characteristics of all languages as ends in them- 
selves. By only examining similar phonetic characteristics in German 
and Welsh, we might tend to consider any common characteristics in the 
physiology and the acoustics to be just as important as any other 
characteristics. Thus, we would miss the notions of function alto- 
gether, notions which indicate that, for example, presence or absence 
of 'voice' serves a contrast ive function in German (see section 7.2) 
while gradual presence of aspiration (or absence of 'voice') serves 
a corresponding contrastive function in Welsh. Without this notion, 
we could simply assign characteristics at random and completely neglect 
the structural importance of these characteristics (compare the binary 
principle, section 6.1). 

The second extreme is to accept the notion of a functional phono- 
logical structure, but, in observing common physiological and acoustic 
characteristics between German and Welsh, to conclude that this similarity 
indicates similarity in structure as well. Such an extreme is most 


prevalent in linguists who, upon observing the universality of physi- 
ology and acoustics, assume a further universality in phonological 
structures. The effect of following this extreme is precisely the same 
as the effect of following the first extreme~we would assign charac- 
teristics to the Welsh phonology on the basis of our knowledge of the 
German phonology, thus neglecting the structural and functional impor- 
tance of the characteristics in Welsh. 

Once we have the physiological and acoustic phonetic data, we must 
determine how the data is to be organized into a structure of phonology, 
or a system of relationships, such that it will reflect the way in which 
the speakers organize the phonetic characteristics in their particular 
language. The problem now is to find a way of determining how a speaker 
organizes the data. 

Some linguists recognize a 'mind' which organizes the data along 
universal rational lines. Unfortunately, the 'knowledge' of this 'mind', 
in whatever form it may take, is not open to examination and cannot be 
verified. On the other hand, we do have evidence that language is a 
function of the brain (see Penfield and Roberts 1959; Whitaker 1971 )» 
so I would first consider neural evidence before considering mental 
or philosophical evidence. With the brain, however, we find that we 
still do not have the evidence necessary to determine the organization 
of a phonology with our current knowledge. Introspection is futile, 
for if we could arrive at the organization of our own phonology through 
introspection (or philosophy), then we could arrive at the organization 
of part of the brain by the same means (indeed, through the same act), 
which we unfortunately cannot do. 

If the 'mind' and the brain are both inaccessible to phonological 


research at this time, how are we to determine the manner in which a 
speaker organizes the phonology? One way is to ask the speaker to judge 
the differences between sounds. This is done in Malecot 1955 in order 
to arrive at various 'strength' scales. Unless the speaker is aware of 
each and every excitation of the aural nerves and is aware of every 
aspect of the physiological configurations used in speech, such judgments 
can only be suspect. For instance, judgments as to relative strength 
of tension will probably be made with reference to the speaker's oral 
musculature, while the actual determining factor may be in or around 
the speaker's larynx. 

Without direct accessibility to the speaker's 'mind' or brain and 
without reliable judgments of the speaker, we have only one area of 
research left. This is the examination (and, where possible, mani- 
pulation) of the speaker's behavior— what the speaker actually utters 
and, where a writing system is available, writes. 

The rationale, then, behind using the speaker's behavior as a 
basis of determining the speaker's phonology involves necessity rather 
than choice t If we cannot determine the cause, then we must judge the 

effect. Through examining the effect, we may come closer to under- 

standing the phenomenon of which the cause is an integral part. 

In using a behavioral approach in determining the organization of 

phonological structures, I should emphasize that we are not using a 

•behavior 1st' approach, such as that of Skinner 1957» the frailties of 

which are exposed in Chomsky 1959. Behaviorism is simply a subset of 

behavioral studies, and a relatively small subset at that. To judge 

any reliance upon behavioral data in terms of the behaviorist notions 

is to give an argument in the extremes which is analogous to the contention 


that a transformational generative approach is basically genetic 
(compare Staats 1971). 

The behavioral evidence (that is, the data supplied by what the 
speaker actually says or writes), gives considerable support to the 
notion of the gradual prosodic opposition of aspiration. This evidence 
has, in fact, already been presented in Chapter 7. But I should briefly 
summarize the evidence in the light of the hierarchical model and the 
phonetic evidence presented subsequent to that chapter. 

Much of the evidence for the opposition deals with the mutation 
system's relationships. In soft mutation, there is an overt loss of 
what is traditionally thought of as aspiration in several of the 
instances. On the other hand, in the aspirate mutation in nonstandard 
applications to consonants, there is the addition of such aspiration. 
In the hierarchical phonological framework, this can be explained as 
a result of the organization, for the addition or loss of aspiration 
results in the realization of the same member of the obstruction opposi- 
tion with a constraint by the higher or lower aspirate opposition member. 
Phonetically, the relationships involved are connected physiologically 
and acoustically through the reference to particular gradual, con- 
staining characteristics. 

In historical lenition and provection, we find basically the same 
sort of phonological and phonetic relationships. In synchronic provec- 
tion, however, the evidence is more compelling, for the speaker uses 
an aspirate or geminate provection form as an equivalent of the same 
member of the obstruction opposition constrained by the next higher degree 
of prosodic aspiration. Again, the phonetic characteristics of this 
relationship are in keeping with the proposed structure. 


As far as spelling is concerned, the older spellings of the fricative 
•liquids' with an h and the development of the spirant mutation with 
the spelling of the 'voiceless aspirated stop' with an h reflect an 
association between the members of the opposition of aspiration which is 
identical to that found in the hierarchical phonology and that supported 
by the phonetic characteristics. 

Thus, the behavior of the speaker is parallel to the phonological 
organization of the hierarchical model in its treatment of the physi- 
ological and acoustic data. If the behavior reflects the actual mental 
or neural organization of the language, then it supports the hypothesis 
that the speaker organizes the phonology such that there may be a 
gradual opposition realized as phonetic aspiration. 

The process of justifying the notion of a gradual opposition of 
aspiration as an obstruction prosody in the hierarchical model has now 
gone the full circle. While we need to justify the phonology through 
the phonetics, we must also justify the phonetics through the phonology. 
If both justifications are sound, then we have a consistent and valid 
system of relationships. In science, consistence and validity with 
respect to theory and observation represent our only path to knowledge. 


J-To be sure, he also mentions the possibility of an overt gesture- 
see below. 

^Compare also the relationship between cause and effect in the use 
of acoustic rather than physiological phonetics in Jakobson and Halle 



Anderson, Henning. 1973. Abduct ive and deductive change. Language 

Anderson, John, and Charles Jones. 197^. Three theses concerning phono- 
logical representation. Journal of Linguistics 10.1-26. 

Anderson, Stephen R. I969. West Scandinavian vowel systems and the 

ordering of phonological rules. Ph.D. dissertation. Cambridge: MIT. 

Anttila, Raimo. 1972. Introduction to historical and comparative 
linguistics. New Yorki Macmillan. 

Atal, B.S., and S.L. Hanauer. 1971. Speech analysis and synthesis by 
linear prediction of the speech wave. Journal of the Acoustical 
Society of America 50.637-55. 

Bell, Alexander G. 1906. The mechanism of speech. New Yorkt Funk and 

Bell, Alexander M. 1899. English visible-speech in twelve lessons. 
Washington! Volta. 

Bloomfield, Leonard. 1933. Language. New York: Holt, Rhinehart and 

Bowen, John T., and T.J. Rhys Jones, i960. Teach yourself Welsh. 
London 1 English Universities Press. 

Broad, David J. 1973. Phonation. Normal aspects of speech, hearing, 
and language, ed. by F.D. Minifie, T.J. Hixon, and F. Williams, 
127-67. Englewood Cliffs: Prentice-Hall. 

Bwrdd Gwybodau Celtaidd. 1942. Orgraff yr iaith Gymraeg [Orthography 
of the Welsh language], ed. by J. Morris Jones. Second edition. 
Caerdydd: Gwasg Prifysgol Cymru. 

Cherry, E.C., Morris Halle, and Roman Jakobson. 1953. Toward the logical 
description of languages in their phonemic aspect. Language 

Chomsky, Noam. 1957. Syntactic structures. The Hague: Mouton. 

Chomsky, Noam. 1959. A review of B.F. Skinner's Verbal Behavior . 
Language 35.26-58. 



Chomsky, Noam. 1964. Current issues in linguistic theory. The structure 
of language, ed. by J. A. Fodor and J.J. Katz, 50-118. Englewood 
Cliffs: Prentice-Hall. 

Chomsky, Noam. 1965. Aspects of the theory of syntax. Cambridge: MIT. 

Chomsky, Noam. 1972. Empirical issues in the theory of transformational 
grammar. Goals of linguistic theory, ed. by S. Peters. Englewood 
Cliffs: Prentice-Hall. 

Chomsky, Noam, and Morris Halle. 1964. Some controversial questions in 
phonological theory. Journal of Linguistics 1.97-138. 

Chomsky, Noam, and Morris Halle. 1968. The sound pattern of English. 
New York: Harper. 

Coker, C., and 0. Fujimura. 1966. Model for specification of the vocal- 
tract area function. Journal of the Acoustical Society of America 

Cooper, F.S., P. Delattre, A.M. Liberman, J.M. Borst, and L.J. Gerstman. 
1952, Some experiments on the perception of synthetic speech 
sounds. Journal of the Acoustical Society of America 24.597-606. 

Cooper. F.S., A.M. Liberman, and J.M. Borst. 195L The interconversion 
of audible and visible patterns of speech. Proceedings of the 
National Academy of Sciences 37.318-25. 

Curtis, James F. 1954. Systematic research in experimental phonetics 3. 
The case for dynamic analysis in acoustic phonetics. Journal of 
Speech and Hearing Disorders 19.147-57. 

Davles, J.J. Glanmor. 1939. The nasal consonants of a Cardiganshire 
dialect. Zeitschrift fur celtische Philologie 20.304-14. 

Delattre. P.C. 1969. Coarticulation and the locus theory. Studia 
Linguistica 23.1-26. 

Delattre, P.C, A.M. Liberman, and F.S. Cooper. 1955. Acoustic loci 
and transitional cues for consonants. Journal of the Acoustical 
Society of America 27.739-73. 

Esau, Helmut. 1973. The generality principle and the goals of phono- 
logical theory. Linguistische Berichte 28.1-22. 

Evans, D. Simon, i960. Gramadeg Cymraeg Canol [A grammar of Middle 
Welsh], Caerdydd: Gwasg Prifysgol Cymru. 

Evans, H. Meurig, and W.0. Thomas. I968. Y geiriadur mawr [The large 
dictionary]. Fourth edition. Llandybie: Llyfrau'r Dryw. 

Fant, G. 1956. On the predictability of formant levels and spectrum 
envelopes from formant frequencies. For Roman Jakobson, 109-20. 
The Hague: Mouton, 


Fant, G. 1959. Acoustic description and classification of phonetic units. 
Ericsson Technics 1. 

Fant, G. i960. Acoustic theory of speech production. The Hague 1 Mouton. 

Fant, G. 1962. Descriptive analysis of the acoustic aspects of speech. 
Logos 5.2-17. 

Fant, G. 1973. Speech sounds and features. Cambridge 1 MIT. 

Fillmore, Charles J. I963. The position of embedding transformations 
in a grammar. Word 19.208-31. 

Firth, J.R. 1948. Sounds and prosodies. Transactions of the Philological 
Society 1948.127-52. 

Fischer- J^rgensen, Eli. 1954. Acoustic analysis of stop consonants. 
Miscellanea Phonetica 2.42-59. 

Foley, James. 1970a. Systematic morphophonology. Mimeo. 

Foley, James. 1970b. Phonological distinctive features. Folia Linguistica 

Fowkes, Robert A. 1949. Initial mutation of loanwords in Welsh. Word 

Fowkes, Robert A. 1959. Phonological aspects of Welsh-English bilingualism. 
General Linguistics 4.23-32. 

Fries, Charles C., and Kenneth L. Pike. 1949. Coexistent phonemic systems. 
Language 25.20-50. 

Fudge, E.C. 1967. The nature of phonological primes. Journal of Linguistics 

Fujimura, Osamu. 1962. Analysis of nasal consonants. Journal of the 
Acoustical Society of America 34.1865-75. 

Fynes-Clinton, O.H. 1913. The Welsh vocabulary of the Bangor district. 

Gardner, H. 1973. The forgotten lesson of Monsieur C. Psychology Today 
(April) 62-7. 

Gelb, I.J. 1963. A study of writing. Chicago » University of Chicago Press. 

Gleason, H.A., Jr. 1961. An introduction to descriptive linguistics. 
Revised edition. New York: Holt, Rhinehart and Winston. 

Griff en, T.D. 1974a. The archiphoneme in generative phonology. Paper 
presented to the Southeastern Conference on Linguistics XI. 

Griff en, T.D. 1974b. The development of Welsh affricates, a change 
through borrowing. Lingua 34.149-65. 


Griff en, T.D. 1975. On describing the cluster prosody. LACUS Forum 

Griffen, T.D. In press a. Lenis initials in Welsh borrowings. To appear 
in Language Sciences. 

Griffen, T.D. In press b. Toward a nonsegmental phonology. To appear 
in Word. 

Halle, Morris. 1959. The sound pattern of Russiant a linguistic and 
acoustical investigation. The Hague* Mouton, 

Halle, Morris, G.W. Hughes, and J. -P. A. Radley. 1957. Acoustic properties 
of stop consonants. Journal of the Acoustical Society of America 

Halle, Morris, and Kenneth N. Stevens. 1964. Speech recognition: a model 
and a program for research. The structure of language, ed, by J. A. 
Fodor and J.J. Katz, 604-12. Englewood Cliffs i Prentice-Hall. 

Hamp, Eric P. 1951. Morphophonemics of the Keltic mutations. Language 
27.230-47. ' 

Hamp, Eric P. 1971. Diachronic and synchronic rules in Welsh. Generative 
studies in historical linguistics, ed. by Maria Tsiapera, 1-8. 
Edmunton: Linguistic Research. 

Harms, Robert T. 1968. Introduction to phonological theory. Englewood 
Cliffs » Prentice-Hall. 

Heinz, J.M., and K.M. Stevens. I96I, On the properties of voiceless 

fricative consonants. Journal of the Acoustical Society of America 

Heinz, J.M., and K.M. Stevens. 1964. On the derivation of area functions 
and acoustic spectra from cineradiographic film. Journal of the 
Acoustical Society of America 36. 1037-8. 

Hjelmslev, Louis. 1961. Prolegomena to a theory of language. Trans. F.J. 
Whitfield. Madison t University of Wisconsin Press. 

Hjelmslev, Louis. 1970. Language, an introduction. Trans. F.J. Whitfield. 
Madison « University of Wisconsin Press. 

Hockett, C.F. 1954. Two models of grammatical description. Word 10.210-31. 

Hockett, C.F. 1955. A manual of phonology. International Journal of 
American Linguistics. Memoir 11. 

Hoenigswald, Henry M. i960. Language change and linguistic reconstruction. 
Chicago! University of Chicago Press. 

Hooper, Joan B. 1973. Aspects of natural generative phonology. Unpublished 
Ph.D. dissertation. Los Angeles » UCLA. 


Householder, R.W. I965. On some recent claims In phonological theory. 
Journal of Linguistics 1.13-34. 

Hyman, Larry. 1970. How concrete is phonology? Language 46.58-75. 

Jackson, Kenneth. 1953. Language and history in early Britain. Edinburgh t 
Edinburgh University Press. 

Jacobs, Roderick A., and Peter S. Rosenbaum. 1968. English transforma- 
tional grammar. Waltham (Massachusetts) t Blaisdell. 

Jakobson, Roman. 1962. Selected writings of Roman Jakobson, vol. 1. 
The Hague t Mouton. [The concept of the sound law and the teleo- 
logical criterion, 1-2 j Principes de phonologie historique, 202-20; 
Remarques sur 1* evolution phonologique du russe comparee a celle 
des autres langues slaves , 7-116.] 

Jakobson, Roman, Gunnar Fant, and Morris Halle. 1952. Preliminaries to 
speech analysis. Cambridge 1 MIT. 

Jakobson, Roman, and Morris Halle. 1962. Phonology and phonetics. 

Selected writings of Roman Jakobson, vol. 1, 465-504. The Hague: 

James, Dan L. 1966. Modern Welsh course. Cardiff t University of Wales 
Press . 

Jones, D.M. 1949. The accent in Modern Welsh. Bulletin of the Board of 
Celtic Studies 13.63-4. 

Jones, R.M. 1963. Gramadeg cenhedlol a graddio'r iaith [Generative 
grammar and grading the language]. Bulletin of the Board of 
Celtic Studies 20.221-35. 

Jones, Robert Owen. 1969. The status of the glottal fricative in the 
Dyffryn Nantlle dialect. Studla Celtica 4.99-109. 

Jones, Robert Owen. 1971. Comparative dialectology. Studla Celtica 

Jones, W.R. 1966. Bilingualism in Welsh education. Cardiff 1 University 
of Wales Press. 

Kent, R.D., and K.L. Moll. 1969. Vocal -tract characteristics of the 
stop cognates. Journal of the Acoustical Society of America 

Kim, C.-W. I965. On the autonomy of the tensity feature in stop clas- 
sification, with special reference to Korean stops. Word 21.339-59. 

Kim, C.-W. I969. A theory of aspiration. Phonetica 21.107-16. 

King, Robert D. 1969. Historical linguistics and generative grammar. 
Englewood Cliffs 1 Prentice-Hall. 


King, Robert D. 1973. Rule insertion. Language 49.551-78. 

Kiparsky, Paul. 1968. Language universale and linguistic change. 
Universals in linguistic theory, ed. by E. Bach and R. Harms, 
170-203. New York « Holt, Rhinehart and Winston. 

Kiparsky, Paul. 1971. Historical linguistics. A survey of linguistic 
science, ed. by W.O. Dingwall, 576-692. College Parkt University 
of Maryland Press. 

Kiparsky, Paul. 1973. Phonological representations. Three dimensions 
of linguistic theory, ed. by 0. Fujimura. Tokyo : TEG. 

Kisseberth, Charles. 1970a. On the functional unity of phonological rules. 
Linguistic Inquiry 1.291-306. 

Kisseberth, Charles. 1970b. Vowel elision in Tonkawa and derivational 
constraints. Studies presented to Robert B. Lees by his students, 
ed. by J.M. Sadock and A.L. Vaneck, 109-37. Edmunton: Linguistic 

Knudsen, John K. 1973. A generative phonological description of Modern 
Standard Welsh. Ph.D. dissertation. Austin: University of Texas. 

Koenig, W., H.K. Dunn, and L.Y. Lacy. 1946. The sound spectrograph. 
Journal of the Acoustical Society of America 17.19-49. 

Korzybski, Alfred. 1933. Science and sanity « an introduction to non- 
Aristotelian systems and general semantics. Lancaster : Science Press. 

Koutsoudas, Andreas, Gerald Sanders, and Craig Noll. 1974. The applica- 
tion of phonological rules. Language 50.1-28. 

Kozhevnikov, V.A., and L.A. Chlstovich. 1964. Speech: articulation 

and perception. Revised. U.S. Department of Commerce, JPRS: 30,543. 

Ladefoged, P. 1962. Elements of acoustic phonetics. Chicago: University 
of Chicago Press, 

Ladefoged, P. 1967. Linguistic phonetics. UCLA Working Papers in 
Linguistics 6, 

Ladefoged, P. 1971. Preliminaries to linguistic phonetics. Chicago: 
University of Chicago Press. 

Lamb, Sydney. 1966a. Prolegomena to a theory of phonology. Language 

Lamb, Sydney. 1966b. Outline of stratificational grammar. Revised 
edition. Washington: Georgetown University Press. 

Lamb, Sydney. 1975. Mutations and relations. LACUS Forum 1.540-57. 


Lehiste, Use, and Gordon E, Peterson. 1961. Transitions, glides and 

diphthongs. Journal of the Acoustical Society of America 33.268-77. 

Lewis, Henry. 1946. Datblygiad yr iaith Gymraeg [Development of the 

Welsh language]. Second edition. Caerdydd: Gwasg Prifysgol Cymru. 

Lewis, Henry, and Holger Pedersen, 1937. A concise comparative Celtic 
grammar. Gottingen: Vandenhoeck and Ruprecht. 

Liberman, A.M. 1970. The grammars of speech and language. Cognitive 
Psychology 1.301-23. 

Liberman, A.M., F.S. Cooper, D.P. Shankweiler, and M. Studdert-Kennedy. 
I967. Perception of the speech code. Psychological Review 74.431-61. 

Liberman, A.M., P. Ingemann, L. Lisker. P. Delattre, and F.S. Cooper. 
1959. Minimal rules for synthesizing speech. Journal of the 
Acoustical Society of America 31.1490-9. 

Lieberman, P. 1967. Intonation, perception, and language. Cambridge 1 MIT. 

Lieberman, P. 1970. Towards a unified phonetic theory. Linguistic 
Inquiry 1.306-22. 

Lindblom, B.E.F., and J.E.F. Sundberg. 1971. Acoustic consequences of 
lip, tongue, jaw, and larynx movement. Journal of the Acoustical 
Society of America 50.1166-79. 

Lisker, Leigh, and Arthur S. Abramson. 1967. Some effects of context 

on voice onset time in English stops. Language and Speech 10.1-28. 

Lockwood, David. 1970. Introduction to stratif icational linguistics. 
New Yorki Hare our t Brace Jovanovich. 

Lockwood, David. 1975. 'Quasi-etymological • and 'natural' phonology as 
two varieties of the same mistake. LACUS Forum 1.446-57. 

Lockwood, David. Ms. On the nature of the phonetic stratal system. 
Preliminary version. Mimeo. 

Lyons, John. 1968. Introduction to theoretical linguistics. Cambridge. 

MacNeilage. P.F. 1970. Motor control of serial ordering of speech. 
Psychological Review 77.182-96. 

Makkai, Adam, and David Lockwood. 1973. Readings in stratificational 
linguistics. University » University of Alabama Press. 

Makkai, Valerie Becker. 1972. Phonological theory, evolution and current 
practice. New Yorki Holt, 

Malecot, Andre. 1955. An experimental study of force of articulation. 
Studia Linguistica 9.35-44. 


Malmberg, Bertil. 1955. The phonetic basis for syllable division. 
Studia Linguistica 9.80-?. 

Malmberg, Bertil. 1963. Phonetics. New Yorkj Dover. 

Martinet, Andre. 1952. Function, structure, and sound change. Word 8.1-32. 

Martinet, Andre. 1955. Economie des changements phonetiques. Berne i Francke. 

Mermelstein, P. 1973. Articulatory model for the study of speech 

production. Journal of the Acoustical Society of America 53.1070-83. 

Mermelstein, P., S. Maeda, and 0. Fujimura. 1971. Description of tongue 
and lip movement in a jaw-based coordinate system. Journal of the 
Acoustical Society of America 49.104. 

Morgan, T.J. 1952. Y treigladau a'u cystrawen [The mutations and their syn- 
tax]. Caerdydd: Gwasg Frifysgol Gymru. 

Morris Jones, John. 1913. A Welsh grammar, historical and comparative, 
phonology and accidence. Oxford. 

Mulder, J.W.F. 1968. Sets and relations in phonology. Oxford. 

Ohman, S.E.G. 1966. Coarticulation in VCV utterances: spectrographs 

measurements. Journal of the Acoustical Society of America 39.151-68. 

Ohman, S.E.G. 1967. Numerical model of coarticulation. Journal of the 
Acoustical Society of America 41. 310-20. 

Ohman, S.E.G., and K.N. Stevens. 1963. Glneradiographic studies of 
speech: procedures and objectives. Journal of the Acoustical 
Society of America 35.1889. 

Palmer, F.R. 1964. Grammatical categories and their phonetic exponents. 
Proceedings of the Ninth International Congress of Linguists. 
Cambridge, Massachusetts. 

Penfield, Wilder, and Lamar Roberts. 1959. Speech and brain-mechanisms. 
New York: Atheneum. 

Perkell, J.S. 1969. Physiology of speech production. Cambridge: MIT. 

Peterson, G.E., and H.L. Barney. 1952. Control methods used in a study of 
the vowels. Journal of the Acoustical Society of America 24.175-84. 

Peterson, G.E., and J.E. Shoup. I966. A physiological theory of 
phonetics. Journal of Speech and Hearing Research 9.5-67. 

Pilch, Herbert. 1957. Morphologie der Nasalmutation in der Umgangsprache 
von Cardiganshire. Lingua 7.269-73. 

Pilch, Herbert. 1958. Das kymrische Lautsystem. Zeitschrift fur vergleichende 
Sprachforschung 75.24-78. 


Pilch, Herbert. 1971. A syntactic study of colloquial Welsh. Studia 
Geltica 7.138-57. 

Postal, Paul M. 1964. Constituent structure « a study of contemporary 
models of syntactic description. The Hague i Mouton. 

Powell, Thomas. 1884. The treatment of English borrowed words in colloquial 
Welsh. Transactions of the Philological Society 1884.335-78. 

Reich, Peter. 1968. Competence, performance, and relational networks. 

Linguistic Automation Progress Report, March 1968, Yale University. 

Robert, Gruffydd. 1927. Griffith Roberts's Welsh grammar. Caerfyrddinj 
Spurrell. [Originally 1567. Dosparth byrr ar a rhann gyntaf i 
ramadeg Cymraeg le cair layer o bynciau anhepcor i vn a chwennychai 
na doedyd y Cymraeg yn dilediaith, nai serif ennu'n lawn. Milan.] 

Roberts, Alan Lloyd. 1973. Anghenion y gynghanedd [Requirements of the 
cynghanedd]. Caerdyddt Gwasg Prifysgol Cymru. 

Roberts, E.W. 1972. An impressionistic-phonetic contribution to the 
issue of feature-onset-time. Glossa 6 »2. 155-79. 

Robins, R.H. 1957. Aspects of prosodic analysis. Proceedings of the 
University of Durham Philosophical Society 1.1-12. 

Sampson, Geoffrey. 1970. On the need for a phonological base. Language 

Sampson, Geoffrey. 1974. Is there a universal phonetic alphabet? 
Language 50.236-59. 

<fe3aussure, Ferdinand. 1959. Course in general linguistics. Trans. W. 
Baskin. New Yorkj Philosophical Library. 

Schane, Sanford A. 1971. The phoneme revisited. Language 47.504-21. 

Schane, Sanford A. 1973. Generative phonology. Englewood Cliffs » 
Prentice-Hall . 

Shibatani, Masayoshi. 1973. The role of surface phonetic constraints in 
generative phonology. Language 49.87-106. 

Skinner, B.F. 1957. Verbal behavior. New Yorkt Appleton-rCentury-Crofts. 

Staats, Arthur W. 1971. Linguistic -mental istic theory versus an explanatory 
S-R learning theory of language development. The ontogenesis of 
grammar, ed. by D.I. Slobin, 103-50. New Yorki Academic Press. 

Strevens, Peter, i960. Spectra of fricative noise in human speech. 
Language and Speech 3.32-49. 

Sturtevant, E.H. 1969. Linguistic change, an introduction to the his- 
torical study of language. [Originally 1917. J Chicago 1 University 
of Chicago Press. 


Sullivan, William J. 1975. Alternation, transformation, realization, 
and stratification revisited. LACUS Forum 1.472-522. 

Sweet, Henry. 1884. Spoken North Welsh. Transactions of the Philological 
Society 1884.409-84. 

Sweet, Henry. 1906. A primer in phonetics. Third edition. Oxford. 

Thomas, Alan. I96I. Ffonemau cytseiniol tafodiaith Dyffryn Wysg 

[Consonantal phonemes of Dyffryn Wysg dialect]. Bulletin of the 
Board of Celtic Studies 19.193-201. 

Thomas, Alan. I966. Systems in Welsh phonology. Studia Celtica 1.93-127. 

Thomas, Alan, I967, Generative phonology in dialectology. Transactions 
of the Philological Society 1967.179-203. 

Thomas, Ceinwen H. 1964. Rhai o nodweddion ffonolegol tafodiaith Nantgarw 
[Some phonological characteristics of Nantgarw dialect]. Bulletin 
of the Board of Celtic Studies 21.18-26. 

Thomas, Ceinwen. I967. Welsh intonation—a preliminary study. Studia 
Celtica 2.8-28. 

Trubetzkoy, N.S. 1969. Principles of phonology. Trans. C.A.B. Baltaxe. 
Berkleyt University of California Press. 

Twaddell, W. Freeman. 1935. On defining the phoneme. Language Monograph 
No. 16. 

Vachek, Josef. 1966. The linguistic school of Prague, Bloomingtonj 
Indiana University Press, 

Venneman, Theo, 1972. Sound change and markedness theory. Linguistic 
change and generative theory, ed. by R.P. Stockwell and R.K.S. 
Macaulay, 1-21. Bloomingtoni Indiana University Press. 

Venneman, Theo, and Peter Ladefoged. 1973. Phonetic features and 
phonological features. Lingua 32.61-74. 

Watkins, T. Arwyn, I96I, Iaithyddiaeth , adweddau ar astudio iaith 
[Linguistics, aspects of studying language], Caerdyddi Gwasg 
Prifysgol Cymru. 

Weinreich, U., W. Labov, and M.I. Herzog. 1968. Empirical foundations 
for a theory of language change. Directions for historical lin- 
guistics, ed. by Lehman and Malkiel, 94-188. Austin: University 
of Texas Press. 

Welch, Doris H. 1975. Latinate English verbs and nouns: a synchronic, 
diachronlc, and panchronic description. LACUS Forum 1.458-71. 

Wells, Rulon S. 1947. Immediate constituents. Language 23.81-117. 


Whitaker, Harry A. 1971. On the representation of language in the 
human brain. Edmunton: Linguistic Research, 

Yngve, Victor H. 1975. The dilemma of contemporary linguistics. 
LAGUS Forum 1.1-16. 

Zwicky, A. 197^+. Taking a false step. Language 50.215-2^. 


Toby D. Griffen was born on May 12, 1946, in Washington D.C. He 
is the youngest son of Captain Ira P. Griffen (USN, Ret.) and Mrs. 
Gladys Griffen (nee Oney). After attending public school in Bethesda, 
Maryland, he enrolled as a cadet at The Citadel, The Military College 
of South Carolina, and graduated in 1968 with a B.A. degree in Modem 
Language and a commission in the U.S. Army Reserve. From here, he 
attended the University of Virginia, earning an M.A. degree in German 
in I969 and studying for one year beyond the degree. In 1970, he 
reported for active duty in the Army, serving as an officer in the 
Military Police at Fort Eustis, Virginia. With his active duty behind 
him, he enrolled in the Program in Linguistics at the University of 
Florida in 1972. 

His publications include 'The development of Welsh affricates, a 
change through borrowing' ( Lingua 34. 149-65) , 'On describing the cluster 
prosody' (LACUS Forum 1.140-7), 'Lenis initials in Welsh borrowings' (to 
appear in Language Sciences ), and 'Toward a nonsegmental phonology' (to 
appear in Word ). Papers presented include 'The archiphoneme in generative 
phonology' (SECOL Xl), 'On describing the cluster prosody' (LACUS I), 
'The interaction and description of phonologies in contact' (SECOL XII), 
and 'Stratificational grammar and an hierarchical phonology' (SECOL XIIl). 


I certify that I have read this study and that in my opinion it 
conforms to acceptable standards of scholarly presentation and is fully 
adequate, in scope and quality, as a dissertation for the degree of 
Doctor of Philosophy. 

Bohdan Saciuk, Chairman 
Associate Professor of Romance 
Languages and Literatures 

I certify that I have read this study and that in my opinion it 
conforms to acceptable standards of scholarly presentation and is fully 
adequate, in scope and quality, as a dissertation for the degree of 
Doctor of Philosophy. 

Chauncey Chu 

Associate Professor of Romance 

Languages and Literatures 

I certify that I have read this study and that in my opinion it 
conforms to acceptable standards of scholarly presentation and is fully 
adequate, in scope and quality, as a dissertation for the degree of 
Doctor of Philosophy. 



Donald Dew 

Associate Professor of Speech 

I certify that I have read this study and that in my opinion it 
conforms to acceptable standards of scholarly presentation and is fully 
adequate, in scope and quality, as a dissertation for the degree of 
Doctor of Philosophy. 


een A. Sullivan 
Assistant Professor of Behavioral 

I certify that I have read this study and that in my opinion it 
conforms to acceptable standards of scholarly presentation and is fully 
adequate, in scope and quality, as a dissertation for the degree of 
Doctor of Philosophy. 

■'^W UAjLm 0- -JL/A^. =■ 

William J. Sullivan", III 
Assistant Professor of Germanic and 
Slavic Languages and Literatures 

This dissertation was submitted to the Graduate Faculty of the Program 
in Linguistics in the College of Arts and Sciences and to the Graduate 
Council, and was accepted as partial fulfillment of the requirements 
for the degree of Doctor of Philosophy. 

June 1975 

uate School 


3 1262 08554 5381